CN108564062A - A kind of island boundary Fast Segmentation Algorithm based on remote sensing image - Google Patents
A kind of island boundary Fast Segmentation Algorithm based on remote sensing image Download PDFInfo
- Publication number
- CN108564062A CN108564062A CN201810389857.5A CN201810389857A CN108564062A CN 108564062 A CN108564062 A CN 108564062A CN 201810389857 A CN201810389857 A CN 201810389857A CN 108564062 A CN108564062 A CN 108564062A
- Authority
- CN
- China
- Prior art keywords
- segmentation
- remote sensing
- sensing image
- island
- algorithm
- 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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/13—Satellite images
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
- G06F18/2135—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on approximation criteria, e.g. principal component analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
- G06V10/267—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion by performing operations on regions, e.g. growing, shrinking or watersheds
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- Multimedia (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Evolutionary Biology (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Astronomy & Astrophysics (AREA)
- Remote Sensing (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Analysis (AREA)
Abstract
The island boundary Fast Segmentation Algorithm based on remote sensing image that the present invention relates to a kind of, the island boundary Fast Segmentation Algorithm include the following steps：Step S1, the island boundary coarse segmentation towards remote sensing image；Step S2, the island boundary high-precision based on island boundary coarse segmentation optimizes.Its advantage is shown：Under the premise of ensureing remote sensing image information amount, wave band number is reduced, computational efficiency is improved, realizes the first segmentation on island boundary；Phenomena such as using the coarse segmentation result on island boundary as initial evolution curve, then carrying out level set calculating, improve computational efficiency, solving low computational efficiency caused by due to remote sensing image area coverage is big, over-segmentation, realizes that the subdivision on island boundary is cut；The feature for having taken into account remote sensing image, when overcoming traditional images partitioning algorithm and directly applying to remote sensing image identification or analysis, existing calculatings time grows the problems such as low with segmentation precision.
Description
Technical field
The present invention relates to remote sensing image technical fields, specifically, being that a kind of island boundary based on remote sensing image is quick
Partitioning algorithm.
Background technology
China possesses the seas under its jurisdiction of nearly 3,000,000 km, and oceanic area is vast and island is numerous so that island becomes profit
With, exploitation ocean main carriers.There are about 6961 island areas in China at 500 square metres or more, and wherein inhabitant island is total
There are 450.Island seagirt is highly susceptible to the influence of the extreme environments such as typhoon.Meanwhile with manually opening in recent years
Hair so that island ecosystem is undergoing violent variation.Therefore, for the research on island there is important strategy to anticipate
Justice and application value.
Remote sensing technology is a kind of remote earth observation means, and the ground of large area can be effectively observed using remote sensing
Object information provides data basis for research island.Remote sensing image has the characteristics that coverage count, real-time, multiband, passes through
Analysis to remote sensing image：On the one hand the real-time based on remote sensing image, can be quickly obtained in island position, boundary and island
The information such as type of ground objects, are generally investigated for island and statistics provides help；On the other hand, being based on remote sensing image has long-term sequence special
Property, by the remote sensing image of comparative analysis different time, the important informations such as sea island resources variation can be probed into.Island is as special
Geographical unit can accelerate the step that China builds ocean power for the research on island.Remote sensing is as important data acquisition
Means provide important data basis for island research.The present invention is for remote sensing image multiband, real-time, large area covering
The features such as, pass through and improve traditional image partition method, it is proposed that island Boundary Extraction towards coarse segmentation and is cut towards subdivision
The research of island Boundary Extraction two parts, the Fast Segmentation of the island boundary information based on remote sensing image is realized, to meet difference
Application demand.
Image segmentation is the important link between pretreatment and image recognition, analysis and understanding, and image segmentation can be higher
Image recognition, analysis and the understanding of layer provide data basis.In recent years, both at home and abroad a large amount of researcher to image partition method into
It has gone a large amount of research, now two class image partition methods has been summarized respectively：1) it is based on the relevant image segmentation side in region
Method；2) it is based on parametric active contour model image partition method.
(1) it is based on the relevant image partition method in region
It is also referred to as similar split plot design based on the relevant image partition method in region, is will have same gray level or identical group
The pixel for knitting structure flocks together, and forms the different zones in image.For such methods, domestic and international researchers carry out
Numerous studies.Xu Jie et al. proposes a kind of improved maximum similarity region fusion interactive image segmentation algorithm, realizes
Fast and accurately segmentation to fuzzy object.Zou Xu China et al. proposes a kind of spectral clustering figure based on improved measuring similarity
As dividing method, it can effectively be partitioned into the target that user specifies.A kind of atlas image of Liu Hong et al. propositions and target image
The weighting method for measuring similarity centered on cutting object, by the village human body brain Magnetic Resonance Zhong Dou nucleocapsid tissue
Split-run test, it was demonstrated that algorithm have higher segmentation precision.Qin Lei et al. is directed to the ripe strawberry image under complex background
Segmentation problem, it is proposed that a kind of ripe strawberry image partitioning algorithm based on RGB color similarity, experiment show that changing segmentation calculates
Method disclosure satisfy that the requirement of ripe strawberry Mechaniaed harvest.The small equality people in field proposes a kind of relative entropy fuzzy C-means clustering point
Algorithm is cut, solves the problems, such as that traditional fuzzy C mean clusters partitioning algorithm can not obtain the detailed information of complicated image.Once it connect virtuous etc.
People proposes a kind of image retrieval algorithm using color cluster segmentation and Shape Feature Extraction, and it is difficult to solve single image feature
Accurately to express the difference problem between image.Liu Chen et al. is in order to solve the cluster result of k means clustering algorithms easily by initial
Center randomly chooses the problem of influence with noise, it is proposed that a kind of improved quick global k means clustering algorithms effectively improve
The calculating accuracy of cluster segmentation algorithm.Zhu's he et al. joint gray level threshold segmentation and outline shape recognition methods, using more
Grade segmentation strategy, realizes the segmentation to river region in SAR image.It opens up Xiao Ming et al. and proposes that one kind being based on least mean-square error
The method for calculating high-low threshold value, this method are more clear effectively the edge for extracting image.Wang Zhi societies et al. are directed to gradation of image
The big feature of difference, it is proposed that a kind of contour feature method for registering based on k- mean clusters segmentation and Morphological scale-space, the party
Method can effectively extract image outline and precisely be registrated.Razali et al. proposes average, intermediate value and Otsu threshold adaptive threshold
Partitioning algorithm is split dental imaging to estimate the age of a people.Wang et al. proposes a kind of based on fuzzy logic
Carrying out image threshold segmentation method obtain the optimal threshold in image segmentation, solve the problems, such as the determination of optimal threshold.
Similar split plot design is a kind of non-supervised classification, and K- means clustering algorithms, fuzzy C-means clustering are calculated in this kind of research
Method Threshold Segmentation Algorithm is algorithms most in use.The studies above persons are in order to obtain more effective information in image, from raising
Algorithm is improved in terms of arithmetic accuracy, the similar partitioning algorithm of tradition and algorithms of different effectively improve segmentation
Precision.
(2) image partition method based on parametric active contour model
Since traditional images dividing method can enhance the noise of image while extracting image boundary, while threshold value selects
Accuracy affected the precision of image segmentation.And the proposition of movable contour model, object boundary extraction problem is converted to
For the optimization problem of energy functional, the variation of contour line can be automatically processed in image segmentation process, is improved at image
The speed and precision of reason, accelerate the development of cutting techniques.
Guo laughs at beautiful et al. for double source CT feature of image and the single movable contour model based on region or boundary of tradition
Deficiency, it is proposed that it is a kind of based on shape of blood vessel constraint movable contour model dividing method.Zhang Zefan et al. is based on GVF-
Fibroid high intensity focused ultrasound Image Automatic Segmentation algorithm of the Snake modellings based on cross entropy and GVF-Snake,
Modified hydrothermal process can effectively be split high intensity focused ultrasound image.Pan changes et al. to solve noise to weak boundary
The influence of segmentation proposes a kind of new, with centainly anti-dry property parametric active contour model based on Harris matrixes.Zhang Ping
Et al. propose a kind of improved geometric active contour model and its image segmentation algorithm based on area information, improve image
The shortcomings that since segmentation result is strong with initial profile curve position.Hu Xuegang et al. in order to solve be based on parameter active contour mould
The shortcomings that type method, proposes a kind of new New image segmentation method based on parametric active contour model based on Snake models.
Wu Chun clevers et al. are by combining small wave converting method, it is proposed that a kind of dynamic direction gradient vector stream mould of improved combination small echo
Type, innovatory algorithm being capable of efficiently Medical Image Segmentations.Rajendran et al. is by combining fuzzy C-means clustering and gradient vector flow
Model is split tumor image, obtains the exact boundary of tumour.Gradient vector flow is generalized to vector by Jaouen et al.
It is worth flow field, and robust variation segmentation is carried out to the four-dimensional image with active-surface.Toth et al. proposes a kind of improved stream
Capable active appearance models algorithm is split extraction to prostate region.Ivanovska et al. propose a kind of extension based on
The Level Set Method of mask plate, algorithm can effectively divide the non-uniform magnetic resonance image of intensity.Ray et al. proposes one
Snake algorithm of the kind based on Dynamic Programming, restrained effectively the clutter on particle object boundary influences Snake models.
In the research based on parametric active contour model, basic thought is to express target side using continuous curve
Edge.Active contour models are in the nature a curve for carrying energy functional, therefore, it is practical for solution energy to solve object boundary process
The process of the minimum value of functional.During minimum value, it often will appear the problems such as calculation amount is larger, and separation calculation effect is poor.
It in the research of above-mentioned scholar, is improved by being combined with other algorithms, to improve the computational efficiency of algorithm.
(3) case study
By the analysis to above-mentioned image partition method, the characteristics of for remote sensing image, existing image partition method exists
There are the following problems in island boundary Fast Segmentation research based on remote sensing image：
(1) remote sensing image is different from traditional digital picture, is one kind of spatial data, has Analysis On Multi-scale Features, i.e.,
It is different to the extraction accuracy of image information under different segmentation precisions.Such as under coarse segmentation requirement, it can be carried based on remote sensing image
Take the lower boundary information of island atural object precision；And in the case where requirement is cut in subdivision, the sea of relatively fining can be extracted based on remote sensing image
Island boundaries information.Existing image segmentation algorithm effectively raises the precision of image segmentation, but its more Information amount is solid
Fixed traditional images, face remote sensing image quick boundary segmentation demand, existing dividing method often exist computational efficiency it is low, point
The problems such as cutting low precision.
(2) remote sensing image is different from traditional digital picture, is one kind of earth observation data, has a wide range of covering
The characteristics of.In traditional image segmentation algorithm, in order to solve the over-segmentation occurred when segmentation, be difficult to divide sunk area etc. and ask
Topic, researchers often using by and the combination of gradation of image information be split calculating.And towards type of ground objects complexity, it covers
The wide remote sensing image data of lid range, the above method is in initial evolution curve far from object boundary, it may appear that the calculating time is long, mistake
Divide phenomenon.
Chinese patent literature：CN201310024820.X, applying date 2013.01.23, patent name are：A kind of island, reef
Water front rapid extracting method.A kind of island or reef coastline rapid obtaining is disclosed, low resolution is built to raw video first
Then image carries out Mean Shift processing, the low resolution image that obtains that treated, is then marked, finally utilizes area
Domain growth method extracts.
A kind of island or reef coastline rapid obtaining of above patent document, effectively to be believed using the color of chromatic image
Breath carries out the extraction in island (reef) coastline, and reduces the image of the noise informations such as wave in image, by Mean Shift algorithms
It is introduced into and, while using the preferable feature of the connectivity in marine site, by way of element marking, improve tidal saltmarsh speed
Degree and effect.The characteristic of Mean Shift algorithms, can make full use of color information, be suitble to carrying for remote sensing image coastline
It takes.But about a kind of computational efficiency height, characteristics of remote sensing image extraction effect is taken into account, realize the skill of island boundary high-precision optimization
Art scheme is then without corresponding open.
In conclusion there is an urgent need for a kind of computational efficiency height, characteristics of remote sensing image extraction effect is taken into account, realizes that island boundary is high-precision
A kind of island boundary fast partition method based on remote sensing image of optimization is spent, and about this island boundary fast partition method
Current then not relevant report.
Invention content
The purpose of the present invention is being directed to deficiency in the prior art, a kind of computational efficiency height is provided, remote sensing image spy is taken into account
Extraction effect is levied, realizes a kind of island boundary fast partition method based on remote sensing image of island boundary high-precision optimization, and
About this island boundary fast partition method.
Another purpose of the present invention is to provide a kind of method of the island boundary fast partition method based on remote sensing image
Flow.
To achieve the above object, the technical solution adopted by the present invention is that：
A kind of island boundary Fast Segmentation Algorithm based on remote sensing image, which is characterized in that the island boundary is quick
Partitioning algorithm includes the following steps：
Step S1, the island boundary coarse segmentation towards remote sensing image.
Step S2, the island boundary high-precision based on island boundary coarse segmentation optimizes.
As a kind of perferred technical scheme, the coarse segmentation in step S1 is adopted to reduce the information redundancy of remote sensing image
With one new image to be split of PCA algorithm constructions, it is defined as Bnew, image pixel sum is M × N, by new remote sensing image
Grey level is defined as L, then L={ 1,2 ..., i ..., l }, and the number of pixels of the i-th gray level of note is ni, then each grey level
The probability of appearance
Based on maximum between-cluster variance algorithm (Otsu algorithms), it is proposed that nonlinear optimization Otsu partitioning algorithms, specific algorithm
It is as follows：
Object function：minε
Constraints：
Wherein,It is the differently other inter-class variance of species, inter-class variance is bigger, the different classes of easier segmentation of atural object；It is different classes of population variance, is the summation of inter-class variance and variance within clusters.
As a kind of perferred technical scheme, the large area coverage property of remote sensing image is directed in step S2, it is proposed that water
Flat set method improves the first of Level Set Method by combining island boundary coarse segmentation as a result, introducing K-means segmentation results
Beginning evolution curve, specific algorithm are as follows：
It is C=c that initial evolution curve, which is arranged,otsuΙckmeans, C meets following equations：
Wherein, Φ (x) be level set function, by combine GAC models and C-V models the advantages of, the level set side of model
Journey can be written as：
Wherein, spf is symbol pressure function (signed pressure force, SPF), and spf indicates as follows：
The codomain of spf functions is [- 1,1].In spf formulas, C1And C2Indicate the average value of brightness of image inside and outside profile.
As a kind of perferred technical scheme, the level set algorithm is that a mobile plane is embedded into higher one
In the function of dimension, the curve in plane is finally obtained by the movement of plane, this curve is image target edge.
As a kind of perferred technical scheme, the gray value of target area is C1, the gray value of background area is C2There is min
(I (x)) ＜ C1, C2＜ max (I (x)), therefore in target areaThat is spf ＞ 0.In background areaThat is spf ＜ 0.
To realize above-mentioned second purpose, the technical solution adopted by the present invention is that：
A kind of island boundary Fast Segmentation Algorithm based on remote sensing image described at least one of basis above-described embodiment,
It is characterized in that, the operating process of the fast partition method includes the following steps：
Step S1, experimental data and tool；
Step S11, Taizhou island water front distribution map based on visual interpretation is considered as with type of ground objects area statistics table " true
Value ", for being compared in terms of segmentation precision and time efficiency to the carried dividing method of the present invention；
Step S2, the island boundary segmentation mistake towards coarse segmentation；
Step S21, it is primarily based on the dimensionality reduction that PCA realizes multiband remote sensing image, a secondary new image is generated and is used for towards thick
The island boundary segmentation of segmentation.Based on the coarse segmentation island boundary space distribution of new image after dimensionality reduction, the present invention is carried non-thread
Property optimization Otsu dividing methods and traditional Otsu, 2D-Otsu, K-means, FCM dividing methods be compared；
Step S22, nonlinear optimization Otsu dividing methods and Otsu, 2D-Otsu, K-means, five kinds of FCM dividing methods
Under different partitioning algorithms, the calculating time of different partitioning algorithms before and after remote sensing image dimensionality reduction is compared；
Step S3, optimize towards the island boundary high-precision that subdivision is cut；
Using improved binaryzation gaussian filtering Level Set Models (I-SBGFRLS) proposed by the present invention to remote sensing image into
Row experiment, has chosen C-V models, SBGFRLS models and I-SBGFRLS models and compares analysis, from efficiency of algorithm and essence
Two aspect of degree carries out contrast experiment；
Step S31, the comparison of computational efficiency；
Step S32, the comparison of computational accuracy；
Step S4, interpretation of result.
As a kind of perferred technical scheme, step S31, in the comparison of computational efficiency, the parameter of C-V models is set as：
λ1=1, λ2=1, Δ t=0.1, μ=1, υ=0；The parameter of SBGFRLS models is set as：α=50, Δ t=1, σ=1；I-
The parameter of SBGFRLS models is set as：α=50, Δ t=1, σ=1.
As a kind of perferred technical scheme, step 32, computational accuracy relatively in, be true using artificial visual interpretation result
Value compares the segmentation essence of C-V models, SBGFRLS models and I-SBGFRLS models in terms of shape, length and area three
Degree.
As a kind of perferred technical scheme, in improved binaryzation gaussian filtering Level Set Models, different α values pair
Segmentation times influence difference, select α to meet real-time segmentation between 100~300 and require.
The invention has the advantages that：
1, a kind of island boundary fast partition method based on remote sensing image of the invention, takes into account the large area of remote sensing image
Observation, quasi real time with the characteristics such as multiband, be to solve when traditional images partitioning algorithm directly applies to remote sensing image identification or divides
When analysis, the existing calculating time grows the problems such as low with segmentation precision, it is proposed that a kind of island boundary based on remote sensing image is quick
Dividing method.This method realizes the island boundary rough segmentation based on remote sensing image first with the Otsu algorithms of nonlinear optimization
It cuts, the inter-class variance of traditional Otsu is substituted using the variance within clusters on island and seawater atural object classification, that is, makes full use of remote sensing shadow
The end condition of the spectral characteristic optimization algorithm of picture improves the segmentation efficiency of remote sensing image, realizes first point of island boundary
It cuts (i.e. coarse segmentation)；Then, using the coarse segmentation result on island boundary as initial evolution curve, binaryzation Gauss filter is optimized
Wave Level Set Models make initial evolution curve close to object boundary under the premise of ensureing segmentation precision, efficiently solve because
Remote sensing image area coverage it is big and caused by phenomena such as computational efficiency is low, over-segmentation, realize the high-precision optimization on island boundary,
With higher practical value.
2, for the multiband of remote sensing image, it is proposed that the coarse segmentation algorithm on island boundary.The coarse segmentation of remote sensing image
The main thought of algorithm is：Under the premise of ensureing remote sensing image information amount, first with PCA algorithms to multiwave remote sensing
Image carries out dimensionality reduction, remote sensing image wave band number is reduced, to improve computational efficiency.
3, it substitutes inter-class variance using variance within clusters and carries out threshold calculations, improve the segmentation efficiency of remote sensing image, realize
The coarse segmentation on island boundary.
4, for the large area coverage property of remote sensing image, it is proposed that the subdivision segmentation method on island boundary.Remote sensing image
Segmenting the main thought of segmentation method is：Using the coarse segmentation result on island boundary as initial evolution curve, level set is then carried out
It calculates.Since the curve that initially develops is close to object boundary, computational efficiency is improved, solves to cause because remote sensing image area coverage is big
Phenomena such as computational efficiency is low, over-segmentation, realize that the subdivision on island boundary is cut.
5, a kind of island boundary fast partition method based on remote sensing image of the invention, has taken into account the feature of remote sensing image
(large area cover and multiband), when overcoming traditional images partitioning algorithm and directly applying to remote sensing image identification or analysis,
The existing calculating time grows the problems such as low with segmentation precision.
Description of the drawings
Fig. 1 is SPF function meaning figures.
Fig. 2 is the island boundary Fast Segmentation flow chart based on remote sensing image.
Fig. 3 is survey region and experimental data schematic diagram.
Fig. 4 is Taizhou island water front distribution map based on visual interpretation.
Fig. 5 is the coarse segmentation island boundary space distribution map based on image after dimensionality reduction.(a)Otus；(b) 2D-Otus；(c)
K-means；(d)FCM；(e) nonlinear optimization Otsu.
Fig. 6 is the time efficiency contrast schematic diagram based on different dividing methods before and after remote sensing image dimensionality reduction.
Fig. 7 is algorithms of different experimental comparison figure.
Fig. 8 is influence schematic diagrames of the α to iterations.
Specific implementation mode
It elaborates below in conjunction with the accompanying drawings to specific implementation mode provided by the invention.
A kind of island boundary Fast Segmentation Algorithm based on remote sensing image.The algorithm main research is as follows：(1) needle
To the multiband of remote sensing image, it is proposed that the rough segmentation segmentation method on island boundary.The part proposes a kind of using non-linear excellent
The Otsu algorithms of change, main thought are：Dimensionality reduction is carried out to multiwave remote sensing image using PCA algorithms, is ensureing remote sensing image
Under the premise of information content, wave band number is reduced, improves computational efficiency；The class of tradition Otsu is substituted using the variance within clusters of atural object classification
Between variance, that is, make full use of the end condition of the spectral characteristic optimization algorithm of remote sensing image, improve the segmentation efficiency of remote sensing image,
Realize the first segmentation (i.e. coarse segmentation) on island boundary.(2) it is directed to the large area coverage property of remote sensing image, it is proposed that island side
The subdivision segmentation method on boundary.The part is input with the coarse segmentation result on island boundary, it is proposed that a kind of improved binaryzation Gauss
Level Set Models are filtered, main thought is：Using the coarse segmentation result on island boundary as initial evolution curve, ensureing segmentation essence
Under the premise of degree, make initial evolution curve close to object boundary, solves to calculate effect caused by due to remote sensing image area coverage is big
Phenomena such as rate is low, over-segmentation realizes the high-precision optimization (i.e. subdivision is cut) on island boundary.
The Fast Segmentation Algorithm includes the following steps：
Step 1：Island boundary coarse segmentation towards remote sensing image
In order to reduce the information redundancy of remote sensing image, using one new image to be split of PCA algorithm constructions, it is defined as
Bnew, image pixel sum be M × N, the grey level of new remote sensing image is defined as L, then L=1,2 ..., i ...,
L }, the number of pixels of the i-th gray level of note is ni, then each grey level occurs probability
By improved tradition Otsu algorithms, quantifies the mathematical relationship between segmentation threshold and information requirement, designed
For nonlinear optimization Otsu partitioning algorithms, specific algorithm is as follows：
Object function：minε
Constraints：
Wherein,It is the differently other inter-class variance of species, inter-class variance is bigger, the different classes of easier segmentation of atural object；It is different classes of population variance, is the summation of inter-class variance and variance within clusters.
Step 2：Island boundary high-precision based on island boundary coarse segmentation optimizes
Level Set Method is that a mobile plane is embedded into the one-dimensional function of higher, final by the movement of plane
The curve in plane is obtained, this curve is image target edge, by combining island boundary coarse segmentation as a result, introducing K-
Means segmentation results, improve the initial evolution curve of Level Set Method, and specific algorithm is as follows：
It is C=c that initial evolution curve, which is arranged,otsuΙckmeans, C meets following equations：
Wherein, Φ (x) be level set function, by combine GAC models and C-V models the advantages of, the level set side of model
Journey can be written as：
Wherein, spf is symbol pressure function (signed pressure force, SPF), and spf indicates as follows：
The codomain of spf functions is [- 1,1].In spf formulas, C1And C2Indicate the average value of brightness of image inside and outside profile.
As shown in Figure 1, the gray value of target area is C1, the gray value of background area is C2There are min (I (x)) ＜ C1, C2
＜ max (I (x)), therefore in target areaThat is spf ＞ 0.In background areaI.e.
Spf ＜ 0.
One, in conjunction with flow chart 2, the present invention is further described：
Island boundary Fast Segmentation based on remote sensing image, two steps progress, are the sea towards coarse segmentation first respectively
Island boundaries are divided；Optimize followed by the island boundary high-precision cut towards subdivision.
Step 1：Based on the island boundary coarse segmentation for improving Otsu
It the characteristics of for the multiwave characteristic of remote sensing image with remote sensing information difference segmentation precision demand, devises towards thick
The island boundary segmentation algorithm of segmentation.Based on maximum between-cluster variance algorithm (Otsu algorithms), it is proposed that Otsu points of nonlinear optimization
Cut algorithm：
1) PCA algorithms are introduced, dimensionality reduction is carried out to the multi-wavelength data of remote sensing image, reduce the information redundancy of remote sensing image
Degree；
2) it is based on minimum value to judge, adds the end condition of partitioning algorithm, improve the meter of coarse segmentation Remote Sensing Image Segmentation
Calculate efficiency.
Step 2：Island boundary subdivision based on improved binaryzation gaussian filtering Level Set Models is cut
For the large area coverage property of remote sensing image, it is proposed that the calculation towards the island boundary high-precision optimization that subdivision is cut
Method：
The island boundary segmentation result towards coarse segmentation is optimized by K-means segmentation results, as first
Beginning evolution boundary inputs；
Edge-stopping function is substituted using SPF functions；
Using C-V model image global informations, binaryzation gaussian filtering Level Set Models are solved to noise-sensitive and boundary
Leakage phenomenon.
Two, illustrate Fast Segmentation side of the present invention with the island boundary fast partition method application example based on remote sensing image
The operating process of method；
(1) experimental data and tool
Experimental data is City of Taizhou marine site remote sensing image, as shown in Figure 3.The remote sensing image is Landsat-8 shadows
As (7 wave bands), shooting time is that shooting time is August in 2013 6；Fig. 3 show remote sensing image medium wave band 4,3,2
Composograph, ranging from ° 52 ', 27 ° 59 ' to 28 ° 39 ' from 121 ° 6 ' to 121, size is 5000*5000 pixel.
Experimental situation is：10 professional versions of Windows, Intel (R) Core (TM) [email protected], RAM
4.00GB, MATLAB R2013a.
Fig. 4 show Taizhou island water front distribution map based on visual interpretation, and atural object classification is sea and island.Table 1 is base
In Taizhou marine site atural object classification area statistics information of visual interpretation.Seawater, island in Taizhou of Zhejiang marine site have been counted in table 1 respectively
The area and number of pixels of small island.
Taizhou marine site atural object classification area statistics of the table 1 based on visual interpretation
Taizhou island water front distribution map based on visual interpretation is considered as " true value " with type of ground objects area statistics table, is used for
The carried dividing method of the present invention is compared in terms of segmentation precision and time efficiency.
(2) the island boundary segmentation process towards coarse segmentation
It is primarily based on the dimensionality reduction that PCA realizes multiband remote sensing image, a secondary new image is generated and is used for the sea towards coarse segmentation
Island boundaries are divided.Fig. 5 is to be put forward the present invention non-linear excellent based on the coarse segmentation island boundary space distribution of new image after dimensionality reduction
Change Otsu dividing methods and traditional Otsu, 2D-Otsu, K-means, FCM dividing methods are compared.Fig. 6 is remote sensing image
The time efficiency comparison diagram of different dividing methods before and after dimensionality reduction.Table 2 is that the coarse segmentation island boundary based on new image after dimensionality reduction carries
The area comparison taken.
Island boundary coarse segmentation area of the table 2 based on new image after dimensionality reduction compares (unit：Pixel)
It can be seen that by Fig. 5, Fig. 6 and table 2：1) equal to the segmentation of new image after dimensionality reduction under 5 kinds of different partitioning algorithms
Similar segmentation result can be obtained, seawater, island can effectively be divided；2) to based on after dimensionality reduction new image it is thick
It is consistent to the divided area result of different atural objects in Otsu, 2D-Otsu, FCM and K-means method in divided area comparison.
With visual interpretation Comparative result, area segmentation result mistake facet product is smaller, and nonlinear optimization Otsu partitioning algorithm mistake facets are accumulated
It is larger；3) after to remote sensing image dimensionality reduction in the calculating time comparison of different partitioning algorithms, the FCM algorithms calculating time is most short, non-
Linear optimization Otsu partitioning algorithms take second place, and K-means algorithms calculate time longest；4) different points before and after to remote sensing image dimensionality reduction
It cuts in the calculating time comparison of algorithm, the time efficiency of nonlinear optimization Otsu algorithms proposed by the present invention improves 84.8%.
(3) optimize towards the island boundary high-precision that subdivision is cut
It is input with the island boundary of coarse segmentation, in the island Boundary Extraction experiment cut towards subdivision, utilizes the present invention
The improved binaryzation gaussian filtering Level Set Models (I-SBGFRLS) proposed test remote sensing image, extract in image
Island boundary.In order to verify the validity of improved binaryzation gaussian filtering Level Set Models, C-V models, SBGFRLS are had chosen
Model and I-SBGFRLS models compare analysis, and contrast experiment is carried out in terms of efficiency of algorithm and precision two.
1) comparison of computational efficiency
Fig. 7 respectively illustrates segmentation result of the experimental data under three kinds of models.In the figure 7, the first behavior initial profile,
Second behavior result of calculation；It is respectively three class models, respectively C-V models, SBGFRLS models and I-SBGFRLS from left to right
Model.Wherein, the parameter of C-V models is set as：λ1=1, λ2=1, Δ t=0.1, μ=1, υ=0；The parameter of SBGFRLS models
It is set as：α=50, Δ t=1, σ=1；The parameter of I-SBGFRLS models is set as：α=50, Δ t=1, σ=1.It is tied by segmentation
Fruit Fig. 7 can be seen that：1) there is over-segmentation phenomenon when dividing to experimental data in C-V models, by island inside points region
Divided；2) SBGFRLS models initial profile and object boundary distance is farther out and I-SBGFRLS model segmentation result phases
Seemingly；3) initial profile of I-SBGFRLS models is close with object boundary, and segmentation result is closer to object boundary.
3 three kinds of parted pattern efficiency comparatives of table
It is compared in terms of calculating time and iterations two respectively in table 3.As can be seen from Table 3：1) it is calculating
In terms of time, when the identical experimental data of processing size, the calculating time longest of C-V models, SBGFRLS models take second place, this
It is most short that the I-SBGFRLS models that invention proposes calculate the time；2) in terms of iterations, when the identical experiment number of processing size
According to when, the iterations of C-V models are most, and SBGFRLS models take second place, I-SBGFRLS models iterations proposed by the present invention
At least；3) since I-SBGFRLS models initial profile proposed by the present invention and object boundary are close, I-SBGFRLS models
The calculating time with iterations be minimum, computational efficiency highest.
2) comparison of computational accuracy
It is true value using Fig. 2 artificial visual interpretation results, from shape, length to compare the segmentation precision of three kinds of models
With the experimental result of three aspect comparison C-V models of area, SBGFRLS models and I-SBGFRLS models.
Table 4 automatically extracts result and human interpretation's result (unit：Pixel)
Table 5 automatically extracts result and human interpretation's result similarity proof list (precision：0.00)
It is to automatically extract result and human interpretation's Comparative result table from table 4, table 5 is to automatically extract result to tie with human interpretation
Fruit similarity proof list.It can be seen that from table 4,5：1) according to similarity proof list as a result, the different model of three classes is to experiment
Data are split, and can obtain accurate segmentation result；2) wherein, C-V models segmentation precision is slightly below other two classes
Partitioning algorithm；3) SBGFRLS models are consistent with I-SBGFRLS model segmentation results, and the separation calculation precision of two class models is compared
C-V models are slightly excellent.
(4) interpretation of result
In improved binaryzation gaussian filtering Level Set Models, different α values influence segmentation times different.Fig. 8 exhibitions
The graph of relation of different α and iterations is shown.α is tested in figure between 50~500 to the influence of iterations.
Fig. 8 is influence curve figures of the different α to iterations, as can be seen from the figure：1) α is bigger, and iterations are in
It takes over to open and gradually decrease；2) with the increase of α, the influence to iterations reduction gradually weakens；3) when α is more than 300, increase
The value of α is almost nil to the image of iterations.In an experiment, in order to improve the efficiency of algorithm, α can be chosen 100~300
Between come meet real-time segmentation require.
A kind of island boundary fast partition method based on remote sensing image of the present invention, the large area for taking into account remote sensing image are seen
It surveys, quasi real time with the characteristics such as multiband, is to solve when traditional images partitioning algorithm directly applies to remote sensing image identification or analyzes
When, the problems such as existing calculating time length is low with segmentation precision, it is proposed that a kind of island boundary based on remote sensing image is quickly divided
Segmentation method.This method realizes the island boundary coarse segmentation based on remote sensing image first with the Otsu algorithms of nonlinear optimization,
The inter-class variance that traditional Otsu is substituted using the variance within clusters on island and seawater atural object classification, that is, make full use of remote sensing image
The end condition of spectral characteristic optimization algorithm improves the segmentation efficiency of remote sensing image, realizes the first segmentation on island boundary (i.e.
Coarse segmentation)；Then, using the coarse segmentation result on island boundary as initial evolution curve, binaryzation gaussian filtering water is optimized
Flat collection model makes initial evolution curve close to object boundary, efficiently solves because of remote sensing under the premise of ensureing segmentation precision
Image area coverage it is big and caused by phenomena such as computational efficiency is low, over-segmentation, realize the high-precision optimization on island boundary, have
Higher practical value.A kind of island boundary fast partition method based on remote sensing image of the present invention, first, for remote sensing shadow
The multiband of picture, it is proposed that the coarse segmentation algorithm on island boundary；Secondly it substitutes inter-class variance using variance within clusters and carries out threshold value
It calculates, improves the segmentation efficiency of remote sensing image, realize the coarse segmentation on island boundary；Then, for the big face of remote sensing image
Product coverage property, it is proposed that the subdivision segmentation method on island boundary.The main thought of the coarse segmentation algorithm of remote sensing image is：Ensureing
Under the premise of remote sensing image information amount, dimensionality reduction is carried out to multiwave remote sensing image first with PCA algorithms, reduces remote sensing image
Wave band number, to improve computational efficiency.The main thought of the subdivision segmentation method of remote sensing image is：By the coarse segmentation knot on island boundary
Then fruit carries out level set calculating as initial evolution curve.Since the curve that initially develops is close to object boundary, improves and calculate effect
Rate, realizes the subdivision on island boundary at phenomena such as solving low computational efficiency caused by due to remote sensing image area coverage is big, over-segmentation
It cuts.
The above is only a preferred embodiment of the present invention, it is noted that for the ordinary skill people of the art
Member, under the premise of not departing from the method for the present invention, can also make several improvement and supplement, these are improved and supplement also should be regarded as
Protection scope of the present invention.
Claims (9)
1. a kind of island boundary Fast Segmentation Algorithm based on remote sensing image, which is characterized in that the island boundary is quickly divided
Algorithm is cut to include the following steps：
Step S1, the island boundary coarse segmentation towards remote sensing image；
Step S2, the island boundary high-precision based on island boundary coarse segmentation optimizes.
2. island boundary according to claim 1 Fast Segmentation Algorithm, which is characterized in that coarse segmentation in step S1 in order to
The information redundancy for reducing remote sensing image is defined as Bnew, image pixel using one new image to be split of PCA algorithm constructions
Sum is M × N, the grey level of new remote sensing image is defined as L, then L={ 1,2 ..., i ..., l }, the i-th gray level of note
Number of pixels is ni, then each grey level occurs probability
Based on maximum between-cluster variance algorithm (Otsu algorithms), it is proposed that nonlinear optimization Otsu partitioning algorithms, specific algorithm are as follows：
Object function：minε
Constraints：
Wherein,It is the differently other inter-class variance of species, inter-class variance is bigger, the different classes of easier segmentation of atural object；It is
Different classes of population variance is the summation of inter-class variance and variance within clusters.
3. island boundary according to claim 1 Fast Segmentation Algorithm, which is characterized in that be directed to remote sensing image in step S2
Large area coverage property, it is proposed that Level Set Method, by combine island boundary coarse segmentation as a result, introduce K-means segmentation
As a result, improving the initial evolution curve of Level Set Method, specific algorithm is as follows：
It is C=c that initial evolution curve, which is arranged,otsuΙckmeans, C meets following equations：
Wherein, Φ (x) is level set function, by the way that in conjunction with the advantages of GAC models and C-V models, the level set equation of model can
It is written as：
Wherein, spf is symbol pressure function (signed pressure force, SPF), and spf indicates as follows：
The codomain of spf functions is [- 1,1].In spf formulas, C1And C2Indicate the average value of brightness of image inside and outside profile.
4. island boundary according to claim 3 Fast Segmentation Algorithm, which is characterized in that the level set algorithm be by
One mobile plane is embedded into the one-dimensional function of higher, and the curve in plane is finally obtained by the movement of plane, this
Curve is image target edge.
5. island boundary according to claim 3 Fast Segmentation Algorithm, which is characterized in that the gray value of target area is C1,
The gray value of background area is C2There are min (I (x)) ＜ C1, C2＜ max (I (x)), therefore in target areaThat is spf ＞ 0.In background areaThat is spf ＜ 0.
6. a kind of according to any island boundary Fast Segmentation Algorithms based on remote sensing image of claim 1-5, feature
It is, the operating process of the fast partition method includes the following steps：
Step S1, experimental data and tool；
Step S11, Taizhou island water front distribution map based on visual interpretation is considered as " true value " with type of ground objects area statistics table,
For being compared in terms of segmentation precision and time efficiency to the carried dividing method of the present invention；
Step S2, the island boundary segmentation process towards coarse segmentation；
Step S21, it is primarily based on the dimensionality reduction that PCA realizes multiband remote sensing image, a secondary new image is generated and is used for towards coarse segmentation
Island boundary segmentation.Based on the coarse segmentation island boundary space distribution of new image after dimensionality reduction, the present invention is carried non-linear excellent
Change Otsu dividing methods and traditional Otsu, 2D-Otsu, K-means, FCM dividing methods are compared；
Step S22, nonlinear optimization Otsu dividing methods and Otsu, 2D-Otsu, K-means, five kinds of differences of FCM dividing methods
Partitioning algorithm under, the calculating time of different partitioning algorithm before and after remote sensing image dimensionality reduction is compared；
Step S3, optimize towards the island boundary high-precision that subdivision is cut；
Remote sensing image is carried out using improved binaryzation gaussian filtering Level Set Models (I-SBGFRLS) proposed by the present invention real
It tests, has chosen C-V models, SBGFRLS models and I-SBGFRLS models and compare analysis, from efficiency of algorithm and precision two
Aspect carries out contrast experiment；
Step S31, the comparison of computational efficiency；
Step S32, the comparison of computational accuracy；
Step S4, interpretation of result.
7. the island boundary Fast Segmentation Algorithm according to claim 6 based on remote sensing image, which is characterized in that step,
In the comparison of S31 computational efficiencies, the parameter of C-V models is set as：λ1=1, λ2=1, Δ t=0.1, μ=1, υ=0；SBGFRLS
The parameter of model is set as：α=50, Δ t=1, σ=1；The parameter of I-SBGFRLS models is set as：α=50, Δ t=1, σ=
1。
8. the island boundary Fast Segmentation Algorithm according to claim 6 based on remote sensing image, which is characterized in that step
32, computational accuracy relatively in, using artificial visual interpretation result be true value, from shape, length and area three in terms of comparison C-V
The segmentation precision of model, SBGFRLS models and I-SBGFRLS models.
9. the island boundary Fast Segmentation Algorithm according to claim 6 based on remote sensing image, which is characterized in that improving
Binaryzation gaussian filtering Level Set Models in, different α values influence segmentation times different, and α is selected to come between 100~300
Meet real-time segmentation to require.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810389857.5A CN108564062A (en) | 2018-04-27 | 2018-04-27 | A kind of island boundary Fast Segmentation Algorithm based on remote sensing image |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810389857.5A CN108564062A (en) | 2018-04-27 | 2018-04-27 | A kind of island boundary Fast Segmentation Algorithm based on remote sensing image |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108564062A true CN108564062A (en) | 2018-09-21 |
Family
ID=63537097
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810389857.5A Pending CN108564062A (en) | 2018-04-27 | 2018-04-27 | A kind of island boundary Fast Segmentation Algorithm based on remote sensing image |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108564062A (en) |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109615595A (en) * | 2018-12-03 | 2019-04-12 | 中国石油大学(华东) | A kind of level set SAR oil spilling extracting method based on bilateral filtering |
| CN109726639A (en) * | 2018-12-07 | 2019-05-07 | 河北工程大学 | A kind of ground object information extraction method based on unsupervised classification technology |
| CN110264484A (en) * | 2019-06-27 | 2019-09-20 | 上海海洋大学 | A kind of improvement island water front segmenting system and dividing method towards remotely-sensed data |
| CN110298848A (en) * | 2019-07-01 | 2019-10-01 | 南京医科大学眼科医院 | The OCT image area of edema boundary extraction method and system inspired based on biology |
| CN110335287A (en) * | 2019-07-15 | 2019-10-15 | 北华航天工业学院 | The extracting method and device of Architectural drawing data |
| CN111754529A (en) * | 2020-06-28 | 2020-10-09 | 南京医科大学眼科医院 | Lattice boltzmann image segmentation method improved by utilizing symbol pressure function |
| CN113095282A (en) * | 2021-04-29 | 2021-07-09 | 中山大学 | Fire grading method, device, equipment and medium for island subareas |
| CN113129248A (en) * | 2021-04-29 | 2021-07-16 | 中山大学 | Island remote sensing image set obtaining method, device, equipment and medium |
| CN113239616A (en) * | 2021-04-29 | 2021-08-10 | 中山大学 | Island fire rescue method and system based on unmanned ship-aircraft cluster cooperation |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1640361A (en) * | 2005-01-06 | 2005-07-20 | 东南大学 | Positive computerized tomography restoration method for multi-phase horizontal set |
| CN104217422A (en) * | 2014-06-03 | 2014-12-17 | 哈尔滨工程大学 | Sonar image detection method of self-adaption narrow-band level set |
| CN106127784A (en) * | 2016-07-01 | 2016-11-16 | 辽宁工程技术大学 | A kind of high-resolution remote sensing image dividing method |
-
2018
- 2018-04-27 CN CN201810389857.5A patent/CN108564062A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1640361A (en) * | 2005-01-06 | 2005-07-20 | 东南大学 | Positive computerized tomography restoration method for multi-phase horizontal set |
| CN104217422A (en) * | 2014-06-03 | 2014-12-17 | 哈尔滨工程大学 | Sonar image detection method of self-adaption narrow-band level set |
| CN106127784A (en) * | 2016-07-01 | 2016-11-16 | 辽宁工程技术大学 | A kind of high-resolution remote sensing image dividing method |
Non-Patent Citations (4)
| Title |
|---|
| DONGMEI HUANG等: "Multi-dimension and multi-granularity segmentation of remote sensing image based on improved Otsu algorithm", 《2017 IEEE 14TH INTERNATIONAL CONFERENCE ON NETWORKING，SENSING AND CONTROL（ICNSC）》 * |
| 孙林等: "基于双符号压力函数的活动轮廓图像分割方法", 《计算机工程与应用》 * |
| 杨勇等: "基于区域 G A C 模型的二值化水平集图像分割算法", 《计算机应用》 * |
| 霍冠英: "《侧扫声呐图像目标分割》", 31 May 2017 * |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109615595A (en) * | 2018-12-03 | 2019-04-12 | 中国石油大学(华东) | A kind of level set SAR oil spilling extracting method based on bilateral filtering |
| CN109726639A (en) * | 2018-12-07 | 2019-05-07 | 河北工程大学 | A kind of ground object information extraction method based on unsupervised classification technology |
| CN110264484B (en) * | 2019-06-27 | 2022-10-18 | 上海海洋大学 | Improved island shoreline segmentation system and segmentation method for remote sensing data |
| CN110264484A (en) * | 2019-06-27 | 2019-09-20 | 上海海洋大学 | A kind of improvement island water front segmenting system and dividing method towards remotely-sensed data |
| CN110298848A (en) * | 2019-07-01 | 2019-10-01 | 南京医科大学眼科医院 | The OCT image area of edema boundary extraction method and system inspired based on biology |
| CN110335287A (en) * | 2019-07-15 | 2019-10-15 | 北华航天工业学院 | The extracting method and device of Architectural drawing data |
| CN111754529A (en) * | 2020-06-28 | 2020-10-09 | 南京医科大学眼科医院 | Lattice boltzmann image segmentation method improved by utilizing symbol pressure function |
| CN113095282A (en) * | 2021-04-29 | 2021-07-09 | 中山大学 | Fire grading method, device, equipment and medium for island subareas |
| CN113239616A (en) * | 2021-04-29 | 2021-08-10 | 中山大学 | Island fire rescue method and system based on unmanned ship-aircraft cluster cooperation |
| CN113239616B (en) * | 2021-04-29 | 2022-05-03 | 中山大学 | Island fire rescue method and system based on unmanned ship-aircraft cluster cooperation |
| CN113095282B (en) * | 2021-04-29 | 2022-05-17 | 中山大学 | Fire grading method, device, equipment and medium for island subareas |
| CN113129248A (en) * | 2021-04-29 | 2021-07-16 | 中山大学 | Island remote sensing image set obtaining method, device, equipment and medium |
| CN113129248B (en) * | 2021-04-29 | 2023-09-12 | 中山大学 | Island remote sensing image set obtaining method, device, equipment and medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108564062A (en) | A kind of island boundary Fast Segmentation Algorithm based on remote sensing image | |
| CN107220980B (en) | A kind of MRI image brain tumor automatic division method based on full convolutional network | |
| CN101546428B (en) | Image fusion of sequence infrared and visible light based on region segmentation | |
| CN101359365B (en) | Iris positioning method based on maximum between-class variance and gray scale information | |
| Kavitha et al. | An efficient approach for brain tumour detection based on modified region growing and neural network in MRI images | |
| CN109389585B (en) | Brain tissue extraction method based on full convolution neural network | |
| Patil et al. | Cancer cells detection using digital image processing methods | |
| CN110120048B (en) | Three-dimensional brain tumor image segmentation method combining improved U-Net and CMF | |
| CN104361582B (en) | Method of detecting flood disaster changes through object-level high-resolution SAR (synthetic aperture radar) images | |
| CN112785617A (en) | Automatic segmentation method for residual UNet rectal cancer tumor magnetic resonance image | |
| CN106408580A (en) | Liver region extraction method based on fuzzy C mean and mean shift | |
| CN114972248A (en) | Attention mechanism-based improved U-net liver tumor segmentation method | |
| Satheesha et al. | Detection of melanoma using distinct features | |
| Mehena et al. | Brain tumor segmentation and extraction of MR images based on improved watershed transform | |
| CN102968779B (en) | Non-subsample Contourlet domain type MRI (Magnetic Resonance Imaging) image enhancing method based on FCM (Fuzzy C-means) cluster | |
| Anandgaonkar et al. | Brain tumor detection and identification from T1 post contrast MR images using cluster based segmentation | |
| CN110111259A (en) | A kind of multisource image anastomosing method based on regional guidance | |
| Sreeja et al. | Detection of tumours in digital mammograms using wavelet based adaptive windowing method | |
| CN109377461A (en) | A kind of breast X-ray image self-adapting enhancement method based on NSCT | |
| Geweid et al. | Efficient tumor detection in medical images using pixel intensity estimation based on nonparametric approach | |
| Ibrahim et al. | Texture and pixel intensity characterization-based image segmentation with morphology and watershed techniques | |
| Sreedhanya et al. | An automatic brain tumor detection and segmentation using hybrid method | |
| Lal et al. | Multi-level fusion of ct and mri brain images for classifying tumor | |
| Devi et al. | Detection of tumours from MRI scans using Segmentation techniques | |
| Lin et al. | Feature selection algorithm for classification of multispectral MR images using constrained energy minimization |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180921 |