CN108717424A - One kind being based on the matched method for searching three-dimension model of breakdown figure - Google Patents
One kind being based on the matched method for searching three-dimension model of breakdown figure Download PDFInfo
- Publication number
- CN108717424A CN108717424A CN201810380046.9A CN201810380046A CN108717424A CN 108717424 A CN108717424 A CN 108717424A CN 201810380046 A CN201810380046 A CN 201810380046A CN 108717424 A CN108717424 A CN 108717424A
- Authority
- CN
- China
- Prior art keywords
- breakdown
- matrix
- matching
- feature
- model
- 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
- G06T17/00—Three dimensional [3D] modelling, e.g. data description of 3D objects
Abstract
The invention discloses one kind being based on the matched method for searching three-dimension model of breakdown figure, including：According to factorization principle, figure matching in pairs is carried out to any two threedimensional model in database, initial matching matrix and breakdown matrix is calculated；Matrixing is carried out to breakdown matrix, new pairs of feature and breakdown figure mating structure is generated, pairs of feature is decomposed according to singular value decomposition method；The matched similarity score of breakdown figure is calculated, final matching matrix is generated according to principle of optimality, the mould of the matching matrix is similarity between threedimensional model two-by-two in database；It is ranked up according to the similarity value between different models, to realize the retrieval of threedimensional model.Present invention effectively prevents in figure matching process because of time for being calculated as bringing affine matrix and spatially compared with macrooperation cost, traditional figure matching process is combined with factorization, threedimensional model whole view feature is efficiently utilized, figure matching recall precision and precision are improved.
Description
Technical field
The present invention relates to three-dimensional model search fields, more particularly to one kind being based on the matched three-dimensional model search of breakdown figure
Method.
Background technology
Compared with traditional two-dimension picture, threedimensional model can be showed in the texture of object, shape, color characteristic etc.
More fully, display form also more horn of plenty and true.As computer vision and a large amount of of vision collecting equipment are popularized,
Threedimensional model will gradually replace traditional two-dimension picture, be applied to the every field of social production life more and more widely,
Such as fields such as CAD (Computer Aided Design, CAD), virtual reality, amusement, medical imagings.It faces
The threedimensional model of magnanimity growth, nowadays one urgent problem to be solved of computer vision field is how effectively to realize and retouch
It states, retrieve, browse three-dimensional data.Existing method for searching three-dimension model can be generally divided into two types：Three based on model
Tie up target retrieval and the objective search method based on view.This mode classification is that different data are used according to them
Type and corresponding method are distinguished.
Figure matching[1]Retrieval is based on technologies such as Digital Image Processing, computer vision and machine learning, by means of calculating
Machine treatment technology carries out the various visual angles view of object in database the process of analysis comparison.Figure matching is led in computer vision
Central role is played in many problems in domain, as shape matches[2], object classification[3], signature tracking[4], symmetry analysis[5]With it is dynamic
It identifies[6].The structural similarity that figure fits through calculate node to node and edge to edge carrys out the matched purpose of implementation model,
However in many practical applications, the characteristic point of the threedimensional model based on view occurs in aggregation type.In this case, traditional
Figure matching search method need calculation scale big and more sparse affine matrix, this be considerably improved the operation of algorithm at
This.
Figure matches the significant challenge that search method faces at present：When the threedimensional model view feature in database is more
When, the calculating cost of affine matrix is higher, the method that then many existing figure matching process use similarity approximate processing, drop
The arithmetic speed of low three-dimensional model search and accuracy, limit practical ranges.
Invention content
The present invention provides one kind being based on the matched method for searching three-dimension model of breakdown figure, effectively prevents figure and matched
Relatively macrooperation cost in journey because of time for being calculated as bringing affine matrix and spatially, by traditional figure matching process and because
Formula decomposition is combined, and efficiently utilizes whole view features of threedimensional model, is improved figure matching effectiveness of retrieval and precision, is referred to
It is described below：
One kind being based on the matched method for searching three-dimension model of breakdown figure, and the method for searching three-dimension model includes following step
Suddenly：
According to factorization principle, figure matching in pairs is carried out to any two threedimensional model in database, is calculated
Initial matching matrix and breakdown matrix；
Matrixing is carried out to breakdown matrix, new pairs of feature and breakdown figure mating structure are generated, according to strange
Different value decomposition method decomposes pairs of feature；
The matched similarity score of breakdown figure is calculated, final matching matrix is generated according to principle of optimality, the matching
Matrix norm is the similarity between threedimensional model two-by-two in database；
It is ranked up according to the similarity value between different models, to realize the retrieval of threedimensional model.
The method further includes：
The various visual angles colored views of each object are acquired, the initial views collection of each object is obtained after extracting mask, carries out feature
Extraction, various visual angles model library is defined as by total view-set of all objects.
The breakdown matrix includes：
The edge connection of node affinity matrix, edge affinity matrix and two threedimensional model graph structures differentiates square
Battle array.
It is described that matrixing is carried out to breakdown matrix, generate new pairs of feature and breakdown figure mating structure
Step is specially：
Wherein, L is pairs of feature, KpFor the node affinity matrix of breakdown figure mating structure, KqIt is matched for breakdown figure
The edge affinity matrix of structure, G1、G2Discrimination matrix is connected for the edge of two threedimensional model graph structures；T is transposition；
It is as follows to generate breakdown figure mating structure：
Wherein, H1、H2For the breakdown figure mating structure of two threedimensional models, I is unit matrix.
Described the step of being decomposed to pairs of feature according to singular value decomposition method is specially：
L=UVT
Wherein, U is horizontal point of vector, and V is vertical point of vector.
The advantageous effect of technical solution provided by the invention is：
1, existing figure matching algorithm needs calculation scale big and more sparse affine matrix, and which greatly enhances calculations
The operation cost of method proposes breakdown figure matching process to solve this problem, in the figure of any pair given of threedimensional model
In structure, with small scale and more dense breakdown matrix replaces original affinity matrix；
2, whole view features of threedimensional model, the matching of breakdown figure is efficiently utilized to preferably resolve in the matching process
The problems such as spilling caused by view feature is more and the approximate convergence during optimization, there is better robustness, improve
The efficiency and precision of three-dimensional model search.
Description of the drawings
Fig. 1 is a kind of flow chart based on the matched method for searching three-dimension model of breakdown figure；
Fig. 2 is the colored views sample of object in University Of Tianjin MV-RED (various visual angles colour-depth) database；
Fig. 3 is the RGB (colour) and depth image for obtaining banana in University Of Tianjin's MV-RED databases from different angles
Sample；
Fig. 4 is that looking into for six kinds of algorithms quasi- looks into full curve；
Fig. 5 is the performance evaluating column diagram of six kinds of algorithms.
Specific implementation mode
To make the object, technical solutions and advantages of the present invention clearer, embodiment of the present invention is made below further
It is described in detail on ground.
Preferably to solve the problems of the prior art, the embodiment of the present invention proposes breakdown figure matching process, given
Any pair of threedimensional model graph structure in, with small scale and more dense breakdown matrix replaces original affinity square
Battle array efficiently utilizes whole view features of threedimensional model, substantially reduces computation complexity.Different from tradition figure matching, breakdown
Figure matches the approximate receipts during preferably resolving the more caused spilling of view feature in the matching process and optimizing
The problems such as holding back has better robustness.
Embodiment 1
A kind of method for searching three-dimension model based on view, referring to Fig. 1, including：View acquisition, views selection, feature carry
It takes and object matching.Wherein, object matching part is the key content of research, main using matched technology is schemed, for solution point
The problem of Xie Shitu is matched proposes that the figure based on factorization matches sorting algorithm, former according to matrixing and singular value decomposition
Reason, with small scale and more dense breakdown matrix replaces original affinity matrix, improves recall precision and accuracy, refers to
It is described below：
101：The various visual angles colored views of each object are acquired, the initial views collection of each object is obtained after extracting mask, are carried out
Feature extraction；
The embodiment of the present invention is defined as various visual angles model by taking Zernike moment characteristics as an example, by total view-set of all objects
Library.
102：According to factorization principle, figure matching in pairs is carried out to any two threedimensional model in database, is calculated
Obtain initial matching matrix X and breakdown matrix；
Wherein, breakdown matrix includes：Node affinity matrix, edge affinity matrix and two three-dimensional model diagrams
The edge of structure connects discrimination matrix.
103：Matrixing is carried out to breakdown matrix, generates new pairs of feature and breakdown figure mating structure, root
Pairs of feature is decomposed according to singular value decomposition method；
104：The matched similarity score of breakdown figure is calculated, final matching matrix is generated according to principle of optimality, it should
The mould of matching matrix is the similarity between threedimensional model two-by-two in database；
105：It is ranked up according to the similarity value between different models, to realize the retrieval of threedimensional model.
In conclusion the embodiment of the present invention through the above steps 101- steps 105 effectively prevent in figure matching process because
It is calculated as the time brought to affine matrix and relatively macrooperation cost spatially, by traditional figure matching process and factorization
It is combined, efficiently utilizes whole view features of threedimensional model, improve figure matching effectiveness of retrieval and precision.
Embodiment 2
201：The various visual angles colored views of each object are acquired, the initial views collection of each object is obtained after extracting mask, are carried out
Total view-set of all objects is defined as threedimensional model various visual angles model library by feature extraction；
One group of view of a threedimensional model M is given, the view of M is expressed asWherein nMIt is M
The number of middle view,It is the single-view view of threedimensional model M, each width view is indicated with Zernike squares.
202：According to factorization principle, figure matching in pairs is carried out to any two threedimensional model in database, is calculated
Obtain initial matching matrix X and breakdown matrix；
In breakdown figure matching process, threedimensional model M is considered as a width figure M={ P, Q, G }, wherein P=[p1,…,pn]
For the node diagnostic vector in graph structure, Q=[q1,…,qm] be graph structure in edge feature vector, G ∈ { 0,1 }n×mFor side
Edge connects discrimination matrix (n is the interstitial content of graph structure, and m is number of edges in graph structure).
Any two threedimensional model M in data-oriented library1、M2, their graph structure is respectively M1={ P1,Q1,G1And M2
={ P2,Q2,G2}.To M1、M2It carries out figure to match to obtain initial matching matrix X, figure is matched according to affinity matrix computational approach
Node and edge in structure carry out affinity calculating respectively, obtain node affinity matrixWith edge affinity
MatrixWherein, R is real number field, n1、n2Number of network nodes in respectively two graph structures, m1、m2Respectively two
Number of edges in graph structure.
Wherein, X is to M1、M2The initial matching matrix that figure matches is carried out, in matrix X, when i-th of node and jth
When a Knot Searching, the element x in matrixij=1, otherwise xij=0.
Node affinity matrix
Edge affinity matrix
Wherein, n1、n2Number of network nodes in respectively two graph structures, m1、m2Number of edges in respectively two graph structures
Mesh,For the node in graph structure,For the edge in graph structure, exp is exponential function, and σ is exponential function
Coefficient.
And the edge connection discrimination matrix G of two threedimensional model graph structures1、G2；Edge connection discrimination matrix G ∈ (0,
1)n×mIn, when node i is connect with node j by edge c, the element g in G matrixic=gcj=1, otherwise gic=gcj=0.
203：The matrix K generated in being matched to breakdown figurep、Kq、G1、G2Matrixing is carried out, new pairs of feature is generated
L, formula are as follows：
Wherein, L is the pairs of feature of breakdown, KpFor the node affinity matrix of breakdown figure mating structure, KqFor breakdown
The edge affinity matrix of figure mating structure, G1、G2Discrimination matrix is connected for the edge of two threedimensional model graph structures；T is to turn
It sets.
L is decomposed according to singular value decomposition method：
L=UVT
Wherein, L is the pairs of feature of breakdown, and U is horizontal point of vector, and V is vertical point of vector.
Next, generating breakdown figure mating structure H1、H2, formula is as follows：
Wherein, H1、H2For the breakdown figure mating structure of two threedimensional models, I is unit matrix.
204：By obtained in above-mentioned steps 202 to 203 as a result, the calculating matched similarity score of breakdown figure, public
Formula is as follows：
Further, by object function Jgm(X) linear combination of a convex function and a concave function is split into, formula is such as
Under：
Jgm(X)=(1- α) Jvex(X)+αJcav(X)
In above formula, convex function component is：
Concave function component is：
To ensure Jgm(X) it is a convex function in maximum everywhere convergent, is iterated for weight coefficient α ∈ [0,1]
Optimization, as α → 0, object function Jgm(X) it levels off to convex function, optimum coefficient α=0.12 is obtained in experiment.
Wherein, JgmFor the matched similarity score of breakdown figure, JvexFor convex function component, JcavFor concave function component, α is
Weight coefficient, | | | |FFor the expression symbol of common norm,For convolution symbol, o is mapping symbols, and vec is vectorization symbol,
Tr is extraction matrix diagonals line element addition and value symbol, and diag is the extraction diagonal of a matrix symbol of element, → approached for variate-value
In symbol.
Principle of optimality is matched according to figure, generates final matching matrix X*, formula is as follows：
Jgm(X) it is a convergent convex function of maximum, to ensure as X=X*, convex function Jgm(X) reach it greatly
At value point, it is as follows that constraint has been carried out to it：
Wherein, X* is the final matching matrix that the matching of breakdown figure generates, JgmFor the matched similarity score of breakdown figure,
n1、n2Number of network nodes in respectively two graph structures,It is n respectively for dimension1、n2It is complete 1 vector,It is for dimension
n1×n2Full 0 matrix.
205：The matching matrix X generated in the algorithm*Mould be the similarity between threedimensional model two-by-two in database；
206：It is ranked up according to the similarity value between different models, to realize the retrieval of 3D models.
In conclusion the embodiment of the present invention through the above steps 201- steps 206 effectively prevent in figure matching process because
It is calculated as the time brought to affine matrix and relatively macrooperation cost spatially, by traditional figure matching process and factorization
It is combined, efficiently utilizes whole view features of threedimensional model, improve figure matching effectiveness of retrieval and precision.
Embodiment 3
Feasibility verification is carried out to the scheme in embodiment 1 with reference to specific experiment, Fig. 2-Fig. 5, it is as detailed below to retouch
It states：
The database that this experiment uses is University Of Tianjin's MV-RED databases, and extremely valuable data base is provided for research
Plinth, comprising 505 objects for belonging to 61 major class in database, the quantitative range of object is differed from 1 to 20 in each category.It is real
It tests and chooses the type comprising no less than 10 target objects as inquiry data, share 311 inquiry targets.721 view structures
At a full version data set, 73 views after sampling constitute a compact version data set.505 objects in database
Data are collected in the real-world object in life.The retrieval of 3D objects is to be based on multi-modal information, i.e. multiple color and depth
View, therefore each complete image data of 3D objects is shown by depth and colored two groups of image record sheets, as shown in Figure 2,3.
Evaluation criteria
1, accuracy rate recall rate curve (PR, Precision-recall).Relationship between it is displaying accuracy and recalls
Key index, can intuitively show the performance of retrieval.Precision and the range for recalling angle value are [0,1].
2, the first rank (FT, First tier).It refers to the recall rate of preceding K relevant matches samples, and wherein K is inquiry object
The radix of place class.
3, second-order (ST, Second tier).It refers to the recall rate of preceding 2K relevant matches samples, and wherein K is inquiry object
The radix of class where body.
4, arest neighbors (NN, Nearest neighbor).It refers to that the highest Matching Model of similarity belongs to inquired mesh
The percentage of target type.
5,F-measure.It is the composite measurement of the accuracy rate and recall rate to the retrieval result of fixed quantity.
6, normalization cumulative gain (DCG, Discounted cumulative gain).It is a kind of statistical method, it
The correlated results for being arranged in front can be given to distribute higher weight.
7, average normalized amendment retrieval ordering (ANMRR[7],Average normalized modified
retrieval rank).It provides corresponding weight come the performance of balancing method according to the arrangement position of object correlation.
This method and following five kinds of methods are compared in experiment：
NN (Nearest neighbor), also known as " the three-dimension object searching algorithm based on arest neighbors "；
HAUS (Hausdorff), also known as " the three-dimension object searching algorithm based on Hausdorff distance "；
AVC[8](A Bayesian 3D Search Engine using Adaptive Views Clustering), again
Claim " using three-dimension object searching algorithm of the view classification based on bayesian criterion is suitable for ".
WBGM[9](Weighted Bipartite Graph Matching), also known as " weighting bipartite graph matching algorithm "；
CCFV[10](Camera Constraint-Free View-Based), also known as " based on view under free-viewing angle
Three-dimension object searching algorithm ".
Experimental result
Looking into for six kinds of algorithms quasi- looks into full curve, performance evaluating column diagram comparison result in MV-RED three-dimensional modeling data storehouses
As shown in Figure 4,5.Look into it is quasi- look into full curve and transverse and longitudinal coordinate institute envelope surface product is bigger, it is more excellent to represent retrieval performance.
As shown in Figure 4, the retrieval performance of this method is apparently higher than NN, HAUS, AVC, WBGM and CCFV algorithm.Wherein WBGM
Algorithm is all based on the matched three-dimensional model searching algorithm of figure, since WBGM algorithms are according only to threedimensional model after cluster with this method
The matching result of partial view feature judge the similitude of searched targets, and this method is in view of three in MV-RED databases
Whole view features of dimension module, matching target have higher characteristic polymorphic, and used match information is far more than WBGM
Algorithm significantly improves retrieval performance.
As shown in Figure 5, the experiment evaluation result of this method is apparently higher than NN, HAUS, AVC, WBGM and CCFV algorithm, verification
The feasibility and superiority of this method.
Bibliography：
[1]P.Foggia,G.Percannella,M.Vento.Graph matching and learning in
pattern recognition in the last 10years[J].Pattern Recog.Artif.Intell.,
vol.28,no.1,pp.1–40,2014.
[2]A.C.Berg,T.L.Berg,J.Malik.Shape matching and object recognition
using low distortion correspondences[J].CVPR,2005.
[3]O.Duchenne,A.Joulin,J.Ponce.A graph-matching kernelfor object
categorization[J].ICCV,2011.
[4]H.Jiang,S.X.Yu,D.R.Martin.Linear scale and
rotationinvariantmatching[J].IEEE Trans,PatternAnal.March,33(7):1339–1355,
2011.
[5]M.Chertok,Y.Keller.Spectral symmetry analysis[J].IEEE Trans,
Pattern Anal,March,32(7):1227–1238,2010.
[6]W.Brendel,S.Todorovic.Learning spatiotemporal graphs of human
activities[J].ICCV,2011.
[7]Henning Muller,WolfgangMuller,David Squire,et al.Performance
evaluation in content-based image retrieval:overview and proposals[J].Pattern
Recogn.Lett.22(5)(2001)593–601.
[8]Ansary T F,DaoudiM,Vandeborre J P.A bayesian 3-d search engine
using adaptive views clustering[J].Multimedia,IEEE Transactions on,2007,9(1):
78-88.
[9]Yue Gao,Qionghai Dai,Meng Wang,et al.3D model retrieval using
weighted bipartite graph matching[J].Signal Processing:Image Communication 26
(2011)39–47.
[10]Yue Gao,Jinhui Tang,Richang Hong,et al.Camera Constraint-Free
View-Based 3-D Object Retrieval[J].IEEE Transactions on image proseccing,
vol.21,no.4,April.2012.
It will be appreciated by those skilled in the art that attached drawing is the schematic diagram of a preferred embodiment, the embodiments of the present invention
Serial number is for illustration only, can not represent the quality of embodiment.
The foregoing is merely presently preferred embodiments of the present invention, is not intended to limit the invention, it is all the present invention spirit and
Within principle, any modification, equivalent replacement, improvement and so on should all be included in the protection scope of the present invention.
Claims (5)
1. one kind being based on the matched method for searching three-dimension model of breakdown figure, which is characterized in that the method for searching three-dimension model
Include the following steps：
According to factorization principle, figure matching in pairs is carried out to any two threedimensional model in database, is calculated initial
Matching matrix and breakdown matrix；
Matrixing is carried out to breakdown matrix, new pairs of feature and breakdown figure mating structure are generated, according to singular value
Decomposition method decomposes pairs of feature；
The matched similarity score of breakdown figure is calculated, final matching matrix is generated according to principle of optimality, the matching matrix
Mould be the similarity between threedimensional model two-by-two in database；
It is ranked up according to the similarity value between different models, to realize the retrieval of threedimensional model.
2. according to claim 1 a kind of based on the matched method for searching three-dimension model of breakdown figure, which is characterized in that institute
The method of stating further includes：
The various visual angles colored views of each object are acquired, the initial views collection of each object is obtained after extracting mask, carries out feature extraction,
Total view-set of all objects is defined as various visual angles model library.
3. according to claim 1 a kind of based on the matched method for searching three-dimension model of breakdown figure, which is characterized in that institute
Stating breakdown matrix includes：
Node affinity matrix：
Edge affinity matrix：
Wherein, n1、n2Number of network nodes in respectively two graph structures, m1、m2Number of edges in respectively two graph structures,For the node in graph structure,For the edge in graph structure, exp is exponential function, and σ is exponential function system
Number；
And the edge connection discrimination matrix G of two threedimensional model graph structures1、G2；In edge connection discrimination matrix G ∈ (0,1)n×m
In, when node i is connect with node j by edge c, the element g in G matrixic=gcj=1, otherwise gic=gcj=0.
4. according to claim 1 or 3 a kind of based on the matched method for searching three-dimension model of breakdown figure, feature exists
In described to carry out matrixing to breakdown matrix, the step of generating new pairs of feature and breakdown figure mating structure has
Body is：
Wherein, L is pairs of feature, KpFor the node affinity matrix of breakdown figure mating structure, KqFor breakdown figure mating structure
Edge affinity matrix, G1、G2Discrimination matrix is connected for the edge of two threedimensional model graph structures；T is transposition；
It is as follows to generate breakdown figure mating structure：
Wherein, H1、H2For the breakdown figure mating structure of two threedimensional models, I is unit matrix.
5. according to claim 4 a kind of based on the matched method for searching three-dimension model of breakdown figure, which is characterized in that institute
Stating the step of being decomposed to pairs of feature according to singular value decomposition method is specially：
L=UVT
Wherein, U is horizontal point of vector, and V is vertical point of vector.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810380046.9A CN108717424B (en) | 2018-04-25 | 2018-04-25 | Three-dimensional model retrieval method based on decomposition type graph matching |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810380046.9A CN108717424B (en) | 2018-04-25 | 2018-04-25 | Three-dimensional model retrieval method based on decomposition type graph matching |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108717424A true CN108717424A (en) | 2018-10-30 |
| CN108717424B CN108717424B (en) | 2021-06-11 |
Family
ID=63899190
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810380046.9A Active CN108717424B (en) | 2018-04-25 | 2018-04-25 | Three-dimensional model retrieval method based on decomposition type graph matching |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108717424B (en) |
Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1527251A (en) * | 2002-12-05 | 2004-09-08 | ���ǵ�����ʽ���� | Method and equipment for searching three-dimensional pattern model data base with empirical 3D shape description |
| US20040249809A1 (en) * | 2003-01-25 | 2004-12-09 | Purdue Research Foundation | Methods, systems, and data structures for performing searches on three dimensional objects |
| CN101350016A (en) * | 2007-07-20 | 2009-01-21 | 富士通株式会社 | Device and method for searching three-dimensional model |
| CN101446980A (en) * | 2008-12-26 | 2009-06-03 | 北京大学 | Tridimensional partial shape match and retrieval method based on color rotation picture |
| CN101593205A (en) * | 2009-06-24 | 2009-12-02 | 清华大学 | Method for searching three-dimension model based on video |
| CN101599077A (en) * | 2009-06-29 | 2009-12-09 | 清华大学 | A kind of method of retrieving three-dimensional objects |
| CN103810747A (en) * | 2014-01-29 | 2014-05-21 | 辽宁师范大学 | Three-dimensional point cloud object shape similarity comparing method based on two-dimensional mainstream shape |
| US20170024896A1 (en) * | 2015-07-21 | 2017-01-26 | IAM Robotics, LLC | Three Dimensional Scanning and Data Extraction Systems and Processes for Supply Chain Piece Automation |
| CN106372224A (en) * | 2016-09-07 | 2017-02-01 | 北京拓扑视景科技有限公司 | Three-dimensional model retrieving method and device |
| CN106484692A (en) * | 2015-08-25 | 2017-03-08 | 北京师范大学 | A kind of method for searching three-dimension model |
| CN106651942A (en) * | 2016-09-29 | 2017-05-10 | 苏州中科广视文化科技有限公司 | Three-dimensional rotation and motion detecting and rotation axis positioning method based on feature points |
| CN107066559A (en) * | 2017-03-30 | 2017-08-18 | 天津大学 | A kind of method for searching three-dimension model based on deep learning |
-
2018
- 2018-04-25 CN CN201810380046.9A patent/CN108717424B/en active Active
Patent Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1527251A (en) * | 2002-12-05 | 2004-09-08 | ���ǵ�����ʽ���� | Method and equipment for searching three-dimensional pattern model data base with empirical 3D shape description |
| US20040249809A1 (en) * | 2003-01-25 | 2004-12-09 | Purdue Research Foundation | Methods, systems, and data structures for performing searches on three dimensional objects |
| CN101350016A (en) * | 2007-07-20 | 2009-01-21 | 富士通株式会社 | Device and method for searching three-dimensional model |
| CN101446980A (en) * | 2008-12-26 | 2009-06-03 | 北京大学 | Tridimensional partial shape match and retrieval method based on color rotation picture |
| CN101593205A (en) * | 2009-06-24 | 2009-12-02 | 清华大学 | Method for searching three-dimension model based on video |
| CN101599077A (en) * | 2009-06-29 | 2009-12-09 | 清华大学 | A kind of method of retrieving three-dimensional objects |
| CN103810747A (en) * | 2014-01-29 | 2014-05-21 | 辽宁师范大学 | Three-dimensional point cloud object shape similarity comparing method based on two-dimensional mainstream shape |
| US20170024896A1 (en) * | 2015-07-21 | 2017-01-26 | IAM Robotics, LLC | Three Dimensional Scanning and Data Extraction Systems and Processes for Supply Chain Piece Automation |
| CN106484692A (en) * | 2015-08-25 | 2017-03-08 | 北京师范大学 | A kind of method for searching three-dimension model |
| CN106372224A (en) * | 2016-09-07 | 2017-02-01 | 北京拓扑视景科技有限公司 | Three-dimensional model retrieving method and device |
| CN106651942A (en) * | 2016-09-29 | 2017-05-10 | 苏州中科广视文化科技有限公司 | Three-dimensional rotation and motion detecting and rotation axis positioning method based on feature points |
| CN107066559A (en) * | 2017-03-30 | 2017-08-18 | 天津大学 | A kind of method for searching three-dimension model based on deep learning |
Non-Patent Citations (3)
| Title |
|---|
| AN-AN LIU 等: "Multi-Modal Clique-Graph Matching for View-Based 3D Model Retrieval", 《IEEE TRANSACTIONS ON IMAGE PROCESSING》 * |
| AN-AN LIU 等: "View-based 3D model retrieval via supervised multi-view feature learning", 《MULTIMEDIA TOOLS AND APPLICATIONS VOLUME 77》 * |
| WEIZHI NIE 等: "View-Based 3D Model Retrieval via Multi-graph Matching", 《NEURAL PROCESSING LETTERS VOLUME 48》 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108717424B (en) | 2021-06-11 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Bai et al. | GIFT: Towards scalable 3D shape retrieval | |
| CN107066559B (en) | Three-dimensional model retrieval method based on deep learning | |
| CN104090972B (en) | The image characteristics extraction retrieved for D Urban model and method for measuring similarity | |
| Li et al. | SHREC'12 Track: Generic 3D Shape Retrieval. | |
| CN105243139B (en) | A kind of method for searching three-dimension model and its retrieval device based on deep learning | |
| CN106897390A (en) | Target precise search method based on depth measure study | |
| CN108875076B (en) | Rapid trademark image retrieval method based on Attention mechanism and convolutional neural network | |
| Sun et al. | A multiscale spectral features graph fusion method for hyperspectral band selection | |
| CN110069656A (en) | A method of threedimensional model is retrieved based on the two-dimension picture for generating confrontation network | |
| CN105205135B (en) | A kind of 3D model retrieval methods and its retrieval device based on topic model | |
| Bhattacharjee et al. | Query adaptive instance search using object sketches | |
| Nie et al. | Convolutional deep learning for 3D object retrieval | |
| CN109920050B (en) | Single-view three-dimensional flame reconstruction method based on deep learning and thin plate spline | |
| Wang et al. | Image retrieval based on exponent moments descriptor and localized angular phase histogram | |
| CN113392244A (en) | Three-dimensional model retrieval method and system based on depth measurement learning | |
| Su et al. | 3d-assisted image feature synthesis for novel views of an object | |
| CN111914912B (en) | Cross-domain multi-view target identification method based on twin condition countermeasure network | |
| Wu et al. | Deep texture exemplar extraction based on trimmed T-CNN | |
| CN105868324B (en) | A kind of multi-angle of view target retrieval method based on hidden state model | |
| CN106951501A (en) | A kind of method for searching three-dimension model based on many figure matchings | |
| CN110070626B (en) | Three-dimensional object retrieval method based on multi-view classification | |
| Wang et al. | Hypergraph based feature fusion for 3-D object retrieval | |
| CN108717424A (en) | One kind being based on the matched method for searching three-dimension model of breakdown figure | |
| Nie et al. | Multi-scale CNNs for 3D model retrieval | |
| CN111414802B (en) | Protein data characteristic extraction method |
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 |