CN108648256A - A kind of gray scale core three-dimensional method for reconstructing based on super dimension - Google Patents
A kind of gray scale core three-dimensional method for reconstructing based on super dimension Download PDFInfo
- Publication number
- CN108648256A CN108648256A CN201810475373.2A CN201810475373A CN108648256A CN 108648256 A CN108648256 A CN 108648256A CN 201810475373 A CN201810475373 A CN 201810475373A CN 108648256 A CN108648256 A CN 108648256A
- Authority
- CN
- China
- Prior art keywords
- dimensional
- pattern
- distance
- dictionary
- block
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T15/00—3D [Three Dimensional] image rendering
- G06T15/005—General purpose rendering architectures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/12—Edge-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Cheminformatics (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computer Graphics (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of gray scale core three-dimensional method for reconstructing based on super dimension.Include the following steps：It proposes the dictionary training algorithm that the super dimension of gray scale is rebuild, two dimension and three-dimensional corresponding multilayer dictionary, the prior information as reconstruction is established using feature extraction, schema extraction and clustering algorithm；The neighborhood matching algorithm that two-value constraint is proposed in reconstruction process, takes the image block of different location different apart from calculation, chooses in dictionary and is used as reconstructed results for matched piece；Reconstruction experiment is carried out to rock core gray level image, reconstructed results are compared with training image, the statistical property of true core image and morphological feature, to prove the validity of algorithm.The gray reconstruction of rock core can retain more raw informations than two-value reconstruction, and this method increases weight computing when carrying out matching block search, improves matched robustness.It ensure that reconstructed results and consistency of the original training image on statistical nature.
Description
Technical field
Innovative gray scale core three-dimensional method for reconstructing more particularly to a kind of gray scale proposed based on super dimension of the invention
The three-dimensional rebuilding method of rock core, belongs to image processing field.
Background technology
In petroleum geology research, core three-dimensional microstructure is to study the basis of rock core Macroscopic physical characteristic.Three dimensions
Word rock core can reflect microstructure among pores in pore-size rank, calculate rock core acoustics, electrology characteristic and simulation seepage flow mistake
Journey is the powerful for analyzing rock core microphysics characteristic.However high-precision core three-dimension image is difficult to directly in Practical Project
It obtains, carries out digital modeling using high-precision two-dimensional image, can effectively rebuild core three-dimension image.
Although being gray level image mostly using the image that imaging device obtains, three-dimensional reconstruction algorithm more at present is
For binary image.Gray level image is subjected to binaryzation when reconstruction, then rebuilds two-value 3-D view.However gray level image
It can reflect more information than bianry image, binaryzation can be carried out after rebuilding gray level image, but the bianry image rebuild can not be extensive
Again at gray level image.In order to more retain raw information, the research for the three-dimensional reconstruction algorithm of gray level image is necessary
's.
Current gray scale rock core algorithm for reconstructing mainly has：The simulation based on cross-correlation function of the propositions such as Tahmasebi is calculated
Method (CCSIM, cross-correlation-based simulation), this method can very well be inherited between layers
Continuity, but random variability between layers not easy to control.And Texture Synthesis, basic thought are with small
Scale texture synthesizes the result texture of large-size as sample.
But the above method does not all account for the true three-dimension structure of rock core, it is by existing true core to surpass dimension to rebuild
Image instructs reconstruction process as prior information, using true three-dimensional image information.The present invention proposes a kind of based on super dimension
Gray scale core three-dimensional method for reconstructing, ensure that the statistical nature of Three-Dimensional Gray core image and the original rock core of two dimension after rebuilding
Image is consistent, and morphological feature keeps similitude.The research project is by project of national nature science fund project《Rock is microcosmic non-
Homogeneous texture three-dimensional image reconstruction and resolution enhancement technology research》(61372174) it subsidizes.
Invention content
It is an object of the invention to a kind of innovative new gray scale core three-dimensional method for reconstructing of proposition, and ensure weight
Rear Three-Dimensional Gray core image is built to be consistent with the original core image of two dimension on statistical nature.
The present invention is achieved through the following technical solutions above-mentioned purpose：
(1) true core gray level image I binaryzations are obtained into Ibw, hole is mutually 1, and rock is mutually 0；Set template size
For n, by true core gray level image I and bianry image IbwIt is divided into the three-dimensional subgraph of one-to-one h × w × n；
(2) feature is extracted to the bottom surface of the three-dimensional subgraph of each gray scale, obtains features；Utilize schema extraction side
Method is with all three-dimensional subgraphs of raster paths scanning, using features as gray scale two-dimensional model, corresponding gray scale 3-D view
Block obtains grayscale mode collection Patternsetgray as gray scale three dimensional pattern, is carried to the bottom surface of the three-dimensional subgraph after binaryzation
Two-value two-dimensional model, the 3-D view block of corresponding two-value is taken to obtain binary pattern collection as two-value three dimensional pattern
Patternsetbw；
(3) set classification number M, using Kmeans clustering algorithms to the gray scale two-dimensional model in Patternsetgray into
Set of patterns is divided into M classes by row clustering；
(4) by Patternsetgray gray scale two-dimensional model and its corresponding gray scale three dimensional pattern,
Patternsetbw two-values two-dimensional model and its two-value three dimensional pattern are corresponded to according to generic to be preserved, the established word of output
Allusion quotation.
(5) for gray scale rock core two dimensional image to be reconstructed, it is overlapped a line every time or a row extract image to be reconstructed
Two-dimensional model { Pattern2d1,Pattern2d2,…,Pattern2di,…,Pattern2dN}；
(6) to i-th of two-dimensional model Pattern2diIn the cluster for calculating M class in the pattern and trained dictionary
The distance D of the heartc={ d1,d2,…di,…,dM, find DcIn minimum value Dmin；
(7) in DminIt is matched in corresponding class, it is assumed that have the dictionary of the p corresponding three-dimensional of two dimension former in such
Son calculates Pattern2diWith the gray scale two-dimensional model distance D of all dictionary atoms in suchbg={ dbg1,dbg2,…
dbgi,…,dbgp, by Pattern2diThe two dimensional image block binaryzation at place, obtains Pattern2dbwi, calculate Pattern2diWith
The distance D of the two-value two-dimensional model of all dictionary atoms in suchbbw={ dbbw1,dbbw2,…dbbwi,…,dbbwp, by Dbg
And DbbwWeighting obtains two-dimensional model distance Db=aDbg+bDbbw；
(8) block that the condition of satisfaction is found in dictionary atom, to DbCarry out ascending order arrangement, the minimum corresponding n of block of distance
× n × n gray scale three dimensional patterns are matched 3-D view block, are sequentially completed the reconstruction to entire three-dimensional core block.
The basic principle of the above method is as follows：
It is a kind of completely new three-dimensional rebuilding method that super dimension, which rebuilds (SDR), and this method has mainly used for reference the oversubscription based on study
The thought of resolution algorithm for reconstructing, using existing true core 3-D view as prior information, study two dimensional image block to three-dimensional
The mapping relations of image block establish corresponding dictionary, when rebuilding, using the information of image to be reconstructed and in prior information
Matched part is rebuild, reconstruct the three-dimensional structure come statistical information and morphosis and original image all compared with connect
Closely.The algorithm has reasonably used existing image resource, and imaging device is obtained 3-D view and Mathematical Modelling Method knot
Close, solve obtain that a large amount of 3-D views are expensive using imaging device and conventional three-dimensional algorithm for reconstructing morphologically with original
Beginning image difference is away from too big disadvantage.This method proposes the super dimension algorithm for reconstructing of gray scale core image under the frame that super dimension is rebuild,
The reconstruction of gray scale core image 2 d-to-3 d is realized, as shown in Figure 1.
In the step (1), using the true core gray scale graphics obtained by CT scan, and two-value is carried out to it
Change obtains corresponding two-value core image.
In the step (2), when establishing dictionary, need to carry out schema extraction to true core image, then according to certain
Criterion preserves two-dimensional model and its corresponding three dimensional pattern information.If template size is n, I (x, y, z) indicates true core three
Tie up structure point (x, y, z) gray value, with the template of n × n to I (x, y, z0) sampling, the template of n × n × n is to I (x, y, z0
+ n) sampling, it is illustrated in figure 25 × 5 two dimension pattern plate and 5 × 5 × 5 three-dimensional template samples on the image, if sampling center
For (x0,y0,z0), then obtain two-dimensional model Pattern2d and its corresponding three dimensional pattern Pattern3d.
Pattern2d=I (x0±n/2,y0±n/2,z0) (1)
Pattern3d=I (x0±n/2,y0±n/2,z0±n)
Fixed form size scans entire true core three-dimensional structure according to raster paths, you can obtains gray level image
The set of patterns Patternset, N of two dimension and its three-dimensional structure indicate the number of template pair.For the continuity of Assured Mode collection
With otherness, center position moves n-1 every time when scanning, between two blocks only there are one the overlapped i.e. adjacent block in face it
Between be overlapped a face, for example first center position is (x when scanning0,y0,z0), then the position of next central point is (x0
+n-1,y0,z0)。
When being rebuild, structural information is focused more on, if directly using the gray value of image as pattern information, is matched
Condition is harsh, easily occurs in the case where match block can not be found in dictionary.The reflection of rock core CT image gray levels is the rock core
Ingredient to the absorbability of X-ray, same rock core ingredient is similar, can with the half-tone information of two dimensional image to be reconstructed into
Row is rebuild, and structural information is more paid close attention to during gray scale rock core is rebuild, it is therefore desirable to extract two dimension structure feature to carry out
Dictionary training and reconstruction.In core image, rock particles and hole have apparent gray difference, have significantly at edge
Gray Level Jump, extraction gradient information can reflect structural information.It proposes to extract First-order Gradient to two dimensional image based on this this method
With second order Gradient Features, using feature as matching condition.
First-order Gradient ▽ f at position (x, y) of gray level image f are defined as with vector：
The image of processing is digital quantity, shown in the regions 3x3 such as Fig. 3 (a) for piece image, z expression gray values,
Carry out approximate partial derivative using size for the template of 3x3 herein, mathematical approach has given below：
Formula (4) can be realized with two templates in Fig. 3 (b) by filtering whole image, be carried herein using the operator
Take First-order Gradient.
Second-order differential can embody the direction of gray scale transformation, and Laplace operator is common Second Order Differential Operator, fixed
Justice is：
Wherein：
For the region of a such as 3x3 of Fig. 3 (a)：
Laplace operator is direction-free operator, is calculated simply, and it is real can be filtered calculating with template and image
It is existing.Use the template such as Fig. 4 to image zooming-out second order gradient in the method, with First-order Gradient together as greyscale two-dimension plots
As feature.
Using features as gray scale two-dimensional model, corresponding gray scale 3-D view block is obtained as gray scale three dimensional pattern
To grayscale mode collection Patternsetgray.Two-value two-dimensional model is extracted to the bottom surface of the three-dimensional subgraph after binaryzation, it is corresponding
The 3-D view block of two-value obtains binary pattern collection Patternsetbw as two-value three dimensional pattern.
In the step (3), in order to keep set of patterns more abundant more complete, larger true core image will be used
Carry out training dictionary, obtain a very big set of patterns, if directly being rebuild using the set of patterns of acquisition as the solution space rebuild
When searching for a similar two-dimensional model in entire set of patterns for each two dimensional image block to be reconstructed and will expend a large amount of
Between, cause reconstruction process very very long, therefore this method proposes the dictionary training method of dictionary classification, improves and rebuild efficiency.It should
Classical Kmeans clustering algorithms are mainly utilized in method, are clustered to two-dimensional model Pattern2d, set of patterns is divided
Pattern for different classes, identical class has similitude, and inhomogeneous pattern differentials are larger, and set of patterns is divided into class θM
In formula be：
θM(Patternseti)=ψ (Patternseti(Pattern2di)) i=1,2 ... N ... N (8)
ψ indicates that clustering algorithm, M indicate classification number in formula.Kmeans algorithms calculate it is simple, can preset in advance using extensive
Sample is divided into cluster centre apart from nearest class by the classification number of cluster by calculating sample at a distance from cluster centre
In.In the dictionary training that the super dimension of gray scale is rebuild, by clustering, help quickly to position two dimension to be reconstructed when rebuilding
Which kind of pattern belongs to, then similar pattern of removal search in such.If cluster number be set as m, by entire dictionary according to
Cluster centre is divided into m sub-spaces, two dimension and its corresponding three dimensional pattern is divided into the class of ownership, as the subspace
The structural representation of dictionary atom, dictionary is as shown in Figure 5.
In the step (4), by Patternsetgray gray scale two-dimensional model and its corresponding gray scale three-dimensional mould
Formula, Patternsetbw two-values two-dimensional model and its two-value three dimensional pattern are corresponded to according to generic and are preserved, and are obtained according to true
The dictionary that rock core is established.
In the step (5), when extracting the two dimensional gray pattern of image to be reconstructed, in order to increase the company between block and block
Continuous property can effectively prevent generating blocking artifact when rebuilding, so overlapping a line or a row obtain between adjacent pattern
{Pattern2d1,Pattern2d2,…,Pattern2di,…,Pattern2dN}；
It is first for each two dimensional gray pattern of image to be reconstructed in order to improve reconstruction efficiency in the step (6)
First search and the nearest class of the pattern distance in trained dictionary.
In the step (7), for being found in step (6) with present mode in nearest class, if there is p
The dictionary atom of a corresponding three-dimensional of two dimension calculates Pattern2diWith the gray scale two dimension mould of all dictionary atoms in such
The distance D of formulabg={ dbg1,dbg2,…dbgi,…,dbgp, by Pattern2diThe two dimensional image block binaryzation at place, obtains
Pattern2dbwi, calculate Pattern2diWith the two-value two-dimensional model distance D of all dictionary atoms in suchbbw=
{dbbw1,dbbw2,…dbbwi,…,dbbwp, by DbgAnd DbbwWeighting obtains two-dimensional model distance Db=aDbg+bDbbw.In this method
In, we are estimated using Euclidean distance to carry out similitude between different mode, two n-dimensional vector x=(x1,…,xn) and y=
(y1,…,yn) between Euclidean distance be：
The similarity for measuring two-dimensional model when reconstruction with Euclidean distance indicates more similar apart from smaller.
In step (8), the block of the condition of satisfaction is found in dictionary atom, specifically for different positions, using such as
Lower method：
I. work as Pattern2diWhen being the pattern of first extraction of image to be reconstructed, to DbCarry out ascending order arrangement, distance
The minimum corresponding n × n of block × n gray scale three dimensional patterns are matched 3-D view block；
Ii. work as Pattern2di(first is removed when the pattern for being the preceding n rows extraction of image to be reconstructed), which is carried out
When reconstruction, n × n × n tri- for having been rebuild with its left neighborhood by n × n × n three dimensional patterns of all atoms in Dictionary of Computing
Tie up the distance D of image blocklg={ dlg1,dlg2,…dlgi,…,dlgp, in Dictionary of Computing atom two-value three dimensional pattern with by left neighbour
The distance D of two-value 3-D view block after the block binaryzation in domainlbw={ dlbw1,dlbw2,…dlbwi,…,dlbwp, by DlgAnd Dlbw
Weighting obtains left neighborhood distance Dl=cDlg+dDlbw, by left neighborhood distance with two-dimensional model is distance weighted obtains new distance D,
D=α Db+βDl, alpha+beta=1 carries out ascending order arrangement to D, is apart from the minimum corresponding n × n of block × n gray scale three dimensional patterns
The 3-D view block matched；
Iii. work as Pattern2diWhen being the pattern of the preceding n row extractions of image to be reconstructed (remove first), to the block into
When row is rebuild, pass through n × n × n three dimensional patterns of all atoms in Dictionary of Computing and neighborhood has been rebuild thereafter n × n × n
The distance D of 3-D view blocku={ du1,du2,…dui,…,dup, in Dictionary of Computing atom two-value three dimensional pattern with by rear neighborhood
Block binaryzation after two-value 3-D view block distance Dubw={ dubw1,dubw2,…dubwi,…,dubwp, by DugAnd DubwAdd
Power obtains rear neighborhood distance Du=eDug+fDubw, by rear neighborhood distance with two-dimensional model is distance weighted obtains new distance D, D
=α Db+γDu,+γ=1 α carries out ascending order arrangement to D, is apart from the minimum corresponding n × n of block × n gray scale three dimensional patterns
Matched 3-D view block；
Iv. work as Pattern2diWhen being the two-dimensional model of rest part extraction of image to be reconstructed, which is rebuild
When, n × n × n graphics for having been rebuild with its left neighborhood by n × n × n three dimensional patterns of all atoms in Dictionary of Computing
As the distance D of block1g=d1g1, d1g2... d1gi..., d1gp, in Dictionary of Computing atom two-value three dimensional pattern with by the block of left neighborhood
The distance D of two-value 3-D view block after binaryzationlbw={ dlbw1, dlbw2... dlbwi..., dlbwp, by DlgAnd DlbwIt weights
To left neighborhood distance Dl=cDlg+dDlbw, n × n × n three dimensional patterns of all atoms have been weighed with neighborhood thereafter in Dictionary of Computing
The distance D of the n × n built up × n 3-D view blocksu={ du1, du2... dui..., dup, two-value is three-dimensional in Dictionary of Computing atom
Two-value 3-D view block distance D of the pattern after the block binaryzation by rear neighborhoodubw={ dubw1, dubw2... dubwi...,
dubwp, by DugAnd DubwWeighting obtains rear neighborhood distance Du=eDug+fDubw, finally with two-dimensional model it is distance weighted obtain it is new
Distance D, D=α Db+βDl+γDu, alpha+beta+γ=1 carries out ascending order arrangement, the minimum corresponding n × n of block of distance × n ashes to D
It is matched 3-D view block to spend three dimensional pattern.
Description of the drawings
Fig. 1 is the super dimension reconstruction process of gray scale rock core
Fig. 2 is two dimension and its three dimensional pattern extraction process
Fig. 3 is the regions image 3x3 and template used herein, wherein (a) is the regions image 3x3；(b) it is used herein one
Rank gradient template
Fig. 4 is the image second order gradient template used herein
Fig. 5 is dictionary structure schematic diagram
Fig. 6 is that super dimension rebuilds Experiment Training image
Fig. 7-1 is the gray-scale map of two-dimentional test image
Fig. 7-2 is the binary map of two-dimentional test image
Fig. 8 is test image gray reconstruction result figure
Fig. 9-1 is test image grey level histogram
Fig. 9-2 is initial three-dimensional structure grey level histogram
Fig. 9-3 is super dimension reconstructed results grey level histogram
Figure 10-1 is to rebuild the directions rock core x two point correlation function
Figure 10-2 is to rebuild the directions rock core y two point correlation function
Figure 10-3 is to rebuild the directions rock core z two point correlation function
Figure 11-1 is to rebuild rock core x dimension linear path functions
Figure 11-2 is to rebuild rock core y dimension linear path functions
Figure 11-3 is to rebuild rock core z dimension linear path functions
Specific implementation mode
With reference to specific embodiments and the drawings, the invention will be further described：
Embodiment：
(1) in this example, schemed using 600 × 600 × 600 gray scale core three-dimensional CT images I of such as Fig. 6 as training
Picture obtains 600 without the certain true core CT images of information and stores in a computer.And by this gray level image hole
It is mutually 1, rock is mutually 0, obtains the core image I after binaryzationbw, template size is set as 3, by true core gray level image I
With bianry image IbwIt is divided into one-to-one 600 × 600 × 3 three-dimensional subgraph；
(2) feature is extracted to the bottom surface of the three-dimensional subgraph of each gray scale, obtains features；Utilize schema extraction side
Method is with all three-dimensional subgraphs of raster paths scanning, using features as gray scale two-dimensional model, corresponding gray scale 3-D view
Block obtains grayscale mode collection Patternsetgray as gray scale three dimensional pattern, is carried to the bottom surface of the three-dimensional subgraph after binaryzation
Two-value two-dimensional model, the 3-D view block of corresponding two-value is taken to obtain binary pattern collection as two-value three dimensional pattern
Patternsetbw；
(3) set classification number M, using Kmeans clustering algorithms to the gray scale two-dimensional model in Patternsetgray into
Set of patterns is divided into M classes by row clustering；
(4) by Patternsetgray gray scale two-dimensional model and its corresponding gray scale three dimensional pattern,
Patternsetbw two-values two-dimensional model and its two-value three dimensional pattern are corresponded to according to generic to be preserved, and dictionary is exported.
(5) 2-D gray image is used to carry out three-dimensional reconstruction, Fig. 7-1 is gray level image to be reconstructed, and Fig. 7-2 is to be reconstructed
Bianry image.Overlapping a line or a row extract the two-dimensional model { Pattern of image to be reconstructed every time2d1,
Pattern2d2..., Pattern2di..., Pattern2dN}；
(6) to i-th of two-dimensional model Pattern2diIn the cluster for calculating M class in the pattern and trained dictionary
The distance D of the heartc={ d1,d2,…di..., dM }, find DcIn minimum value Dmin；
(7) in DminIt is matched in corresponding class, it is assumed that have the dictionary of the p corresponding three-dimensional of two dimension former in such
Son calculates Pattern2diWith the gray scale two-dimensional model distance D of all dictionary atoms in suchbg={ dbg1,dbg2,…
dbgi,…,dbgp, by Pattern2diThe two dimensional image block binaryzation at place, obtains Pattern2dbwi, calculate Pattern2diWith
The distance D of the two-value two-dimensional model of all dictionary atoms in suchbbw={ dbbw1,dbbw2,…dbbwi,…,dbbwp, by Dbg
And DbbwWeighting obtains two-dimensional model distance Db=aDbg+bDbbw；
(8) block of the condition of satisfaction is found in dictionary atom according to following criterion：
I. work as Pattern2diWhen being the pattern of first extraction of image to be reconstructed, to DbCarry out ascending order arrangement, distance
Minimum corresponding 3 × 3 × 3 gray scale three dimensional pattern of block is matched 3-D view block；
Ii. work as Pattern2di(first is removed when the pattern for being the preceding n rows extraction of image to be reconstructed), which is carried out
When reconstruction, rebuild with its left neighborhood by 3 × 3 × 3 three dimensional patterns of all atoms in Dictionary of Computing 3 × 3 × 3 three
Tie up the distance D of image blocklg={ dlg1,dlg2,…dlgi,…,dlgp, in Dictionary of Computing atom two-value three dimensional pattern with by left neighbour
The distance D of two-value 3-D view block after the block binaryzation in domainlbw={ dlbw1,dlbw2,…dlbwi,…,dlbwp, by DlgAnd Dlbw
Weighting obtains left neighborhood distance Dl=cDlg+dDlbw, by left neighborhood distance with two-dimensional model is distance weighted obtains new distance D,
D=α Db+βDl, alpha+beta=1 carries out ascending order arrangement to D, is apart from minimum corresponding 3 × 3 × 3 gray scale three dimensional pattern of block
The 3-D view block matched；
Iii. work as Pattern2diWhen being the pattern of the preceding n row extractions of image to be reconstructed (remove first), to the block into
When row is rebuild, pass through 3 × 3 × 3 three dimensional patterns of all atoms in Dictionary of Computing and neighborhood has been rebuild thereafter n × n × n
The distance D of 3-D view blocku={ du1,du2,…dui,…,dup, in Dictionary of Computing atom two-value three dimensional pattern with by rear neighborhood
Block binaryzation after two-value 3-D view block distance Dubw={ dubw1,dubw2,…dubwi,…,dubwp, by DugAnd DubwAdd
Power obtains rear neighborhood distance Du=eDug+fDubw, by rear neighborhood distance with two-dimensional model is distance weighted obtains new distance D, D
=α Db+γDu,+γ=1 α carries out ascending order arrangement to D, is apart from minimum corresponding 3 × 3 × 3 gray scale three dimensional pattern of block
Matched 3-D view block；
Iv. work as Pattern2diWhen being the two-dimensional model of rest part extraction of image to be reconstructed, which is rebuild
When, 3 × 3 × 3 graphics rebuild with its left neighborhood by 3 × 3 × 3 three dimensional patterns of all atoms in Dictionary of Computing
As the distance D of blocklg={ dlg1,dlg2,…dlgi,…,dlgp, in Dictionary of Computing atom two-value three dimensional pattern with by left neighborhood
The distance D of two-value 3-D view block after block binaryzationlbw={ dlbw1,dlbw2,…dlbwi,…,dlbwp, by DlgAnd DlbwWeighting
Obtain left neighborhood distance Dl=cDlg+dDlbw, n × n × n three dimensional patterns of all atoms have been weighed with neighborhood thereafter in Dictionary of Computing
The distance D of the n × n built up × n 3-D view blocksu={ du1,du2,…dui,…,dup, two-value is three-dimensional in Dictionary of Computing atom
Two-value 3-D view block distance D of the pattern after the block binaryzation by rear neighborhoodubw={ dubw1,dubw2,…dubwi,…,
dubwp, by DugAnd DubwWeighting obtains rear neighborhood distance Du=eDug+fDubw, finally with two-dimensional model it is distance weighted obtain it is new
Distance D, D=α Db+βDl+γDu, alpha+beta+γ=1 carries out ascending order arrangement, corresponding 3 × 3 × 3 ash of the minimum block of distance to D
It is matched 3-D view block to spend three dimensional pattern.
(9) according to the criterion of step (8), the three-dimensional reconstruction of whole picture core image, reconstructed results such as Fig. 8 institutes are sequentially completed
Show.
(10) it in geostatistics, is described usually using statistical natures such as two point correlation function, linear path functions
Interstitial space.Therefore, usually by comparing reconstructed results and image to be reconstructed and initial three-dimensional structure in three-dimensional reconstruction algorithm
Statistical eigenfunction verify the validity of reconstructed results.In the method we using two point correlation function and in advance
Path function verifies the validity of this method.
It can be seen that the reconstructed results of this method have continuity in a z-direction from the reconstructed results of Fig. 8.Such as Fig. 9-1
It is test image grey level histogram, Fig. 9-2 is initial three-dimensional structure grey level histogram, and Fig. 9-3 is that super dimension reconstructed results gray scale is straight
Fang Tu.Can effectively rebuild gray level image from the above-mentioned it can be seen from the figure that algorithm, the grey level histogram of reconstructed results with it is to be reconstructed
Image, the grey level histogram of original three-dimensional image are consistent.By image to be reconstructed, the three-dimensional structure of initial three-dimensional structure and reconstruction
Binaryzation, then calculates respective two point correlation function and linear path function, and Figure 10-1 is to rebuild two point phase of the directions rock core x
Function is closed, Figure 10-2 is to rebuild the directions rock core y two point correlation function, and Figure 10-3 is to rebuild the directions rock core z two point correlation function,
Figure 11-1 is to rebuild rock core x dimension linear path functions, and Figure 11-2 is to rebuild rock core y dimension linear path functions, Figure 11-3
It is to rebuild rock core z dimension linear path functions.It can be seen that the two-point probability function of this paper algorithm reconstructed results from above-mentioned figure
It is similar to initial three-dimensional structure, image to be reconstructed with linear path function, it was demonstrated that the statistical nature of the algorithm reconstructed results with it is true
Real gray scale core three-dimension image is consistent, can obtain and preferable rebuild effect, it was demonstrated that the validity of this method.
Above-described embodiment is presently preferred embodiments of the present invention, is not the limitation to technical solution of the present invention, as long as
Without the technical solution that creative work can be realized on the basis of the above embodiments, it is regarded as falling into of the invention special
In the rights protection scope of profit.
Claims (2)
1. a kind of gray scale core three-dimensional method for reconstructing based on super dimension, it is characterised in that：Include the following steps：
(1) true core gray level image I binaryzations are obtained into Ibw, hole is mutually 1, and rock is mutually 0；Template size is set as n, is incited somebody to action
True core gray level image I and bianry image IbwIt is divided into the three-dimensional subgraph of one-to-one h × w × n；
(2) feature is extracted to the bottom surface of the three-dimensional subgraph of each gray scale, obtains features；Using schema extraction method with light
All three-dimensional subgraphs of grid path scanning, using features as gray scale two-dimensional model, corresponding gray scale 3-D view block conduct
Gray scale three dimensional pattern obtains grayscale mode collection Patternsetgray, and two-value is extracted to the bottom surface of the three-dimensional subgraph after binaryzation
The 3-D view block of two-dimensional model, corresponding two-value obtains binary pattern collection Patternsetbw as two-value three dimensional pattern；
(3) classification number M is set, the gray scale two-dimensional model in Patternsetgray is clustered using Kmeans clustering algorithms
Analysis, is divided into M classes by set of patterns；
(4) by Patternsetgray gray scale two-dimensional model and its corresponding gray scale three dimensional pattern, Patternsetbw bis-
It is worth two-dimensional model and its two-value three dimensional pattern and corresponds to preservation, the established dictionary of output according to generic；
(5) for gray scale rock core two dimensional image to be reconstructed, it is overlapped a line every time or a row extract the two dimension of image to be reconstructed
Pattern { Pattern2d1,Pattern2d2,…,Pattern2di,…,Pattern2dN}；
(6) to i-th of two-dimensional model Pattern2diCalculate the pattern and the cluster centre of M class in trained dictionary
Distance Dc={ d1,d2,…di,…,dM, find DcIn minimum value Dmin；
(7) in DminIt is matched in corresponding class, it is assumed that the dictionary atom for having the p corresponding three-dimensional of two dimension in such calculates
Pattern2diWith the gray scale two-dimensional model distance D of all dictionary atoms in suchbg={ dbg1,dbg2,…dbgi,…,
dbgp, by Pattern2diThe two dimensional image block binaryzation at place, obtains Pattern2dbwi, calculate Pattern2diWith in such
The distance D of the two-value two-dimensional model of all dictionary atomsbbw={ dbbw1,dbbw2,…dbbwi,…,dbbwp, by DbgAnd DbbwWeighting
Obtain two-dimensional model distance Db=aDbg+bDbbw；
(8) block that the condition of satisfaction is found in dictionary atom, to DbCarry out ascending order arrangement, the minimum corresponding n × n of the block × n of distance
Gray scale three dimensional pattern is matched 3-D view block, is sequentially completed the reconstruction to entire three-dimensional core block.
2. the super dimension method for reconstructing of gray scale rock core according to claim 1, it is characterised in that：
In the step (1), using the true core gray scale graphics obtained by CT scan, and binaryzation is carried out to it and is obtained
Corresponding two-value core image；
In the step (2), when establishing dictionary, need to carry out schema extraction to true core image, then according to certain criterion
Preserve two-dimensional model and its corresponding three dimensional pattern information.If template size is n, I (x, y, z) indicates true core three-dimensional structure
In the gray value of point (x, y, z), with the template of n × n to I (x, y, z0) sampling, the template of n × n × n is to I (x, y, z0+ n) it adopts
Sample, is illustrated in figure 25 × 5 two dimension pattern plate and 5 × 5 × 5 three-dimensional template samples on the image, if sampling center is (x0,
y0,z0), then obtain two-dimensional model Pattern2d and its corresponding three dimensional pattern Pattern3d.
Pattern2d=I (x0±n/2,y0±n/2,z0)
Pattern3d=I (x0±n/2,y0±n/2,z0±n)
For the continuity and otherness of Assured Mode collection, center position moves n-1 every time when scanning, between two blocks only
It is overlapped a face between the overlapped i.e. adjacent block in one face, for example first center position is (x when scanning0,y0,z0), then
The position of next central point is (x0+n-1,y0,z0)。
Patternset={ (Pattern2d1,Pattern3d1),…(Pattern2di,Pattern3di),…
(Pattern2dN,Pattern3dN)}
Structural information is more paid close attention to during carrying out gray scale rock core reconstruction, it is therefore desirable to extract two dimension structure feature to carry out word
Allusion quotation training and reconstruction, in core image, rock particles and hole have apparent gray difference, have apparent gray scale at edge
Saltus step, extraction gradient information can reflect structural information, propose to extract First-order Gradient and two to two dimensional image based on this this method
Rank Gradient Features, by First-order Gradient GORO DAIMON gradient together as greyscale two-dimension plots as feature；
Using features as gray scale two-dimensional model, corresponding gray scale 3-D view block obtains ash as gray scale three dimensional pattern
Set of patterns Patternsetgray is spent, two-value two-dimensional model, corresponding two-value are extracted to the bottom surface of the three-dimensional subgraph after binaryzation
3-D view block as two-value three dimensional pattern, obtain binary pattern collection Patternsetbw；
In the step (3), in order to improve reconstruction efficiency, this method proposes the dictionary training method of dictionary classification, this method master
Classical Kmeans clustering algorithms are utilized, two-dimensional model Pattern2d is clustered, set of patterns is divided into different
The pattern of class, identical class has similitude, and inhomogeneous pattern differentials are larger, and set of patterns is divided into class θMIn formula
For：
θM(Patternseti)=ψ (Patternseti(Pattern2di)) i=1,2 ... N
ψ indicates that clustering algorithm, M indicate classification number in formula.Kmeans algorithms calculate it is simple, can preset in advance cluster using extensive
Classification number, by calculate sample at a distance from cluster centre, sample is divided into cluster centre in nearest class,
In the dictionary training that the super dimension of gray scale is rebuild, by clustering, help quickly to position two-dimensional model category to be reconstructed when rebuilding
In which kind of, then similar pattern of removal search in such, if cluster number is set as m, by entire dictionary according in cluster
The heart is divided into m sub-spaces, and two dimension and its corresponding three dimensional pattern are divided into the class of ownership, and the dictionary as the subspace is former
Son；
In the step (4), by Patternsetgray gray scale two-dimensional model and its corresponding gray scale three dimensional pattern,
Patternsetbw two-values two-dimensional model and its two-value three dimensional pattern are corresponded to according to generic to be preserved, and is obtained according to true core
The dictionary established；
In the step (5), when extracting the two dimensional gray pattern of image to be reconstructed, in order to increase the continuity between block and block,
It can effectively prevent generating blocking artifact when rebuilding, so overlapping a line or a row obtain between adjacent pattern
{Pattern2d1,Pattern2d2,…,Pattern2di,…,Pattern2dN}；
In the step (6), in order to improve reconstruction efficiency, for each two dimensional gray pattern of image to be reconstructed, exist first
Search and the nearest class of the pattern distance in trained dictionary；
In the step (7), for being found in step (6) with present mode in nearest class, if there is p two dimension
The dictionary atom of corresponding three-dimensional calculates Pattern2diWith the gray scale two-dimensional model of all dictionary atoms in such away from
From Dbg={ dbg1,dbg2,…dbgi,…,dbgp, by Pattern2diThe two dimensional image block binaryzation at place, obtains
Pattern2dbwi, calculate Pattern2diWith the two-value two-dimensional model distance D of all dictionary atoms in suchbbw={ dbbw1,
dbbw2,…dbbwi,…,dbbwp, by DbgAnd DbbwWeighting obtains two-dimensional model distance Db=aDbg+bDbbw.In the method, we
It is estimated using Euclidean distance to carry out similitude between different mode, two n-dimensional vector x=(x1,…,xn) and y=(y1,…,yn)
Between Euclidean distance be：
The similarity for measuring two-dimensional model when reconstruction with Euclidean distance indicates more similar apart from smaller；
In the step (8), the block of the condition of satisfaction is found in dictionary atom, specifically for different positions, using as follows
Method：
I. work as Pattern2diWhen being the pattern of first extraction of image to be reconstructed, ascending order arrangement is carried out to Db, distance minimum
Corresponding n × the n of block × n gray scale three dimensional patterns are matched 3-D view block；
Ii. work as Pattern2di(first is removed when the pattern for being the preceding n rows extraction of image to be reconstructed), which is rebuild
When, n × n × n 3-D views for having been rebuild with its left neighborhood by n × n × n three dimensional patterns of all atoms in Dictionary of Computing
The distance D of blocklg={ dlg1,dlg2,…dlgi,…,dlgp, two-value three dimensional pattern and the block two by left neighborhood in Dictionary of Computing atom
The distance D of two-value 3-D view block after valuelbw={ dlbw1,dlbw2,…dlbwi,…,dlbwp, by DlgAnd DlbwWeighting obtains
Left neighborhood distance Dl=cDlg+dDlbw, by left neighborhood distance with two-dimensional model is distance weighted obtains new distance D, D=α Db+βDl,
Alpha+beta=1 carries out ascending order arrangement to D, and the minimum corresponding n × n of the block × n gray scale three dimensional patterns of distance are matched graphics
As block；
Iii. work as Pattern2di(first is removed when the pattern for being the preceding n row extractions of image to be reconstructed), which is rebuild
When, pass through n × n × n three dimensional patterns of all atoms in Dictionary of Computing and neighborhood has been rebuild thereafter n × n × n 3-D views
The distance D of blocku={ du1,du2,…dui,…,dup, two-value three dimensional pattern and the block two-value by rear neighborhood in Dictionary of Computing atom
The distance D of two-value 3-D view block after changeubw={ dubw1,dubw2,…dubwi,…,dubwp, by DugAnd DubwAfter weighting obtains
Neighborhood distance Du=eDug+fDubw, by rear neighborhood distance with two-dimensional model is distance weighted obtains new distance D, D=α Db+γDu, α
+ γ=1 carries out ascending order arrangement to D, and the minimum corresponding n × n of the block × n gray scale three dimensional patterns of distance are matched graphics
As block；
Iv. work as Pattern2diWhen being the two-dimensional model of rest part extraction of image to be reconstructed, when being rebuild to the block, lead to
Cross n × n × n three dimensional patterns of all atoms in Dictionary of Computing and n × n × n 3-D view blocks that its left neighborhood has been rebuild
Distance Dlg={ dlg1,dlg2,…dlgi,…,dlgp, two-value three dimensional pattern and the block binaryzation by left neighborhood in Dictionary of Computing atom
The distance D of two-value 3-D view block afterwardslbw={ dlbw1,dlbw2,…dlbwi,…,dlbwp, by DlgAnd DlbwWeighting obtains left neighbour
Domain distance Dl=cDlg+dDlbw, n × n × n three dimensional patterns of all atoms and neighborhood has been rebuild thereafter n in Dictionary of Computing ×
The distance D of n × n 3-D view blocksu={ du1,du2,…dui,…,dup, in Dictionary of Computing atom two-value three dimensional pattern with will after
The distance D of two-value 3-D view block after the block binaryzation of neighborhoodubw={ dubw1,dubw2,…dubwi,…,dubwp, by DugWith
DubwWeighting obtains rear neighborhood distance Du=eDug+fDubw, finally with two-dimensional model is distance weighted obtains new distance D, D=α Db+
βDl+γDu, alpha+beta+γ=1 carries out ascending order arrangement to D, is apart from the minimum corresponding n × n of block × n gray scale three dimensional patterns
The 3-D view matched.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810475373.2A CN108648256B (en) | 2018-05-17 | 2018-05-17 | Grayscale rock core three-dimensional reconstruction method based on super-dimension |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810475373.2A CN108648256B (en) | 2018-05-17 | 2018-05-17 | Grayscale rock core three-dimensional reconstruction method based on super-dimension |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108648256A true CN108648256A (en) | 2018-10-12 |
| CN108648256B CN108648256B (en) | 2021-07-02 |
Family
ID=63756693
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810475373.2A Active CN108648256B (en) | 2018-05-17 | 2018-05-17 | Grayscale rock core three-dimensional reconstruction method based on super-dimension |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108648256B (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111833432A (en) * | 2019-04-22 | 2020-10-27 | 四川大学 | Three-dimensional reconstruction method based on core two-dimensional gray scale image |
| CN112017274A (en) * | 2019-05-29 | 2020-12-01 | 四川大学 | Multi-resolution three-dimensional core pore fusion method based on pattern matching |
| CN112037124A (en) * | 2019-06-04 | 2020-12-04 | 中国石油化工股份有限公司 | Feature-adjustable digital core reconstruction method based on image texture synthesis |
| CN112785695A (en) * | 2021-02-07 | 2021-05-11 | 安徽医科大学 | Method for constructing three-dimensional simulation model of fatty liver tissue |
| CN113409193A (en) * | 2021-06-18 | 2021-09-17 | 北京印刷学院 | Super-resolution reconstruction method and device for hyperspectral image |
| CN113515847A (en) * | 2021-05-12 | 2021-10-19 | 中国矿业大学 | Heterogeneous rock digital core modeling method based on K-means clustering algorithm |
| CN114764843A (en) * | 2021-01-12 | 2022-07-19 | 四川大学 | Porous medium image super-dimensional reconstruction method based on neighborhood block matching |
| WO2023134310A1 (en) * | 2022-01-11 | 2023-07-20 | 大连理工大学 | High-throughput microstructure characterization and reconstruction method for heterogeneous material |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102087676A (en) * | 2010-12-13 | 2011-06-08 | 上海大学 | Pore network model (PNM)-based bionic bone scaffold designing method |
| US20140002617A1 (en) * | 2012-06-27 | 2014-01-02 | The Board Of Trustees Of The University Of Illinois | Particle tracking system and method |
| CN105006018A (en) * | 2015-06-30 | 2015-10-28 | 四川大学 | Three-dimensional CT core image super-resolution reconstruction method |
| CN105354873A (en) * | 2015-09-18 | 2016-02-24 | 四川大学 | Mode density function simulation algorithm for three-dimensional reconstruction of porous media |
| CN106296653A (en) * | 2016-07-25 | 2017-01-04 | 浙江大学 | Brain CT image hemorrhagic areas dividing method based on semi-supervised learning and system |
| CN107341848A (en) * | 2017-07-08 | 2017-11-10 | 青岛科技大学 | A kind of new method by rock core CT image procossings for commercial CFD code readable documents |
-
2018
- 2018-05-17 CN CN201810475373.2A patent/CN108648256B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102087676A (en) * | 2010-12-13 | 2011-06-08 | 上海大学 | Pore network model (PNM)-based bionic bone scaffold designing method |
| US20140002617A1 (en) * | 2012-06-27 | 2014-01-02 | The Board Of Trustees Of The University Of Illinois | Particle tracking system and method |
| CN105006018A (en) * | 2015-06-30 | 2015-10-28 | 四川大学 | Three-dimensional CT core image super-resolution reconstruction method |
| CN105354873A (en) * | 2015-09-18 | 2016-02-24 | 四川大学 | Mode density function simulation algorithm for three-dimensional reconstruction of porous media |
| CN106296653A (en) * | 2016-07-25 | 2017-01-04 | 浙江大学 | Brain CT image hemorrhagic areas dividing method based on semi-supervised learning and system |
| CN107341848A (en) * | 2017-07-08 | 2017-11-10 | 青岛科技大学 | A kind of new method by rock core CT image procossings for commercial CFD code readable documents |
Non-Patent Citations (2)
| Title |
|---|
| XIAOHAI HE ET AL.: "Similarity Evaluation of 3D Gray Rock Image Using Pattern Density Classification Function", 《ADVANCES IN INTELLIGENT SYSTEMS RESEARCH》 * |
| 夏晨木，等: "岩石三维图像裂缝提取方法", 《计算机工程与应用》 * |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111833432A (en) * | 2019-04-22 | 2020-10-27 | 四川大学 | Three-dimensional reconstruction method based on core two-dimensional gray scale image |
| CN112017274A (en) * | 2019-05-29 | 2020-12-01 | 四川大学 | Multi-resolution three-dimensional core pore fusion method based on pattern matching |
| CN112017274B (en) * | 2019-05-29 | 2022-11-11 | 四川大学 | Multi-resolution three-dimensional core pore fusion method based on pattern matching |
| CN112037124A (en) * | 2019-06-04 | 2020-12-04 | 中国石油化工股份有限公司 | Feature-adjustable digital core reconstruction method based on image texture synthesis |
| CN114764843A (en) * | 2021-01-12 | 2022-07-19 | 四川大学 | Porous medium image super-dimensional reconstruction method based on neighborhood block matching |
| CN114764843B (en) * | 2021-01-12 | 2023-04-18 | 四川大学 | Porous medium image super-dimensional reconstruction method based on neighborhood block matching |
| CN112785695A (en) * | 2021-02-07 | 2021-05-11 | 安徽医科大学 | Method for constructing three-dimensional simulation model of fatty liver tissue |
| CN113515847A (en) * | 2021-05-12 | 2021-10-19 | 中国矿业大学 | Heterogeneous rock digital core modeling method based on K-means clustering algorithm |
| CN113515847B (en) * | 2021-05-12 | 2023-09-05 | 中国矿业大学 | Heterogeneous rock digital core modeling method based on K-means clustering algorithm |
| CN113409193A (en) * | 2021-06-18 | 2021-09-17 | 北京印刷学院 | Super-resolution reconstruction method and device for hyperspectral image |
| CN113409193B (en) * | 2021-06-18 | 2023-07-04 | 北京印刷学院 | Super-resolution reconstruction method and device for hyperspectral image |
| WO2023134310A1 (en) * | 2022-01-11 | 2023-07-20 | 大连理工大学 | High-throughput microstructure characterization and reconstruction method for heterogeneous material |
| US11915386B2 (en) | 2022-01-11 | 2024-02-27 | Dalian University Of Technology | High-throughput microstructure characterization and reconstruction method of heterogeneous materials |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108648256B (en) | 2021-07-02 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108648256A (en) | A kind of gray scale core three-dimensional method for reconstructing based on super dimension | |
| Bai et al. | Liver tumor segmentation based on multi-scale candidate generation and fractal residual network | |
| Ju et al. | Computing a family of skeletons of volumetric models for shape description | |
| CN106611413A (en) | Image segmentation method and system | |
| CN109034045A (en) | A kind of leucocyte automatic identifying method based on convolutional neural networks | |
| CN109978037A (en) | Image processing method, model training method, device and storage medium | |
| Gao et al. | On combining morphological component analysis and concentric morphology model for mammographic mass detection | |
| CN109191471A (en) | Based on the pancreatic cell image partition method for improving U-Net network | |
| Baniukiewicz et al. | Generative adversarial networks for augmenting training data of microscopic cell images | |
| Tan et al. | Analysis of segmentation of lung parenchyma based on deep learning methods | |
| CN108010013A (en) | A kind of lung CT image pulmonary nodule detection methods | |
| CN110349170A (en) | A kind of full connection CRF cascade FCN and K mean value brain tumor partitioning algorithm | |
| Zollhöfer et al. | GPU based ARAP Deformation using Volumetric Lattices. | |
| Chen et al. | Generative adversarial U-Net for domain-free medical image augmentation | |
| CN113269724A (en) | Fine-grained cancer subtype classification method | |
| CN113393546B (en) | Fashion clothing image generation method based on clothing type and texture pattern control | |
| Zhu et al. | Nonlocal low-rank point cloud denoising for 3-D measurement surfaces | |
| Shao et al. | Application of U-Net and Optimized Clustering in Medical Image Segmentation: A Review. | |
| CN112489048B (en) | Automatic optic nerve segmentation method based on depth network | |
| CN111833432B (en) | Three-dimensional reconstruction method based on core two-dimensional gray scale image | |
| CN110992320B (en) | Medical image segmentation network based on double interleaving | |
| Wang et al. | PointFilterNet: A filtering network for point cloud denoising | |
| Chang | Research on sports video image based on fuzzy algorithms | |
| Yan et al. | Two and multiple categorization of breast pathological images by transfer learning | |
| CN111127408B (en) | Automatic delineation method and system for clinical target area of primary focus of nasopharyngeal carcinoma based on GrowCut |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |