CN109829505A - A kind of threedimensional model classification method based on various features descriptor - Google Patents
A kind of threedimensional model classification method based on various features descriptor Download PDFInfo
- Publication number
- CN109829505A CN109829505A CN201910123038.0A CN201910123038A CN109829505A CN 109829505 A CN109829505 A CN 109829505A CN 201910123038 A CN201910123038 A CN 201910123038A CN 109829505 A CN109829505 A CN 109829505A
- Authority
- CN
- China
- Prior art keywords
- threedimensional model
- feature
- feature vector
- various features
- point
- 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
Landscapes
- Image Analysis (AREA)
Abstract
The invention discloses a kind of threedimensional model classification methods based on various features descriptor, comprising: selects various features descriptor to extract the feature vector of threedimensional model, feature descriptor includes Gaussian curvature, shape diameter function and scale invariant feature conversion；Denoising dimension-reduction treatment is carried out to feature vector using Robust Principal Component Analysis；It is clustered using density-based algorithms, obtains threedimensional model classification results.The present invention, which controls basic data using various features descriptor, can avoid unicity, and the feature vector denoising dimensionality reduction obtain to extraction simultaneously clustered using density-based algorithms, can get the result of more accurate classification.
Description
Technical field
The present invention relates to threedimensional model sorting technique fields, and in particular to a kind of three-dimensional mould based on various features descriptor
Type classification method.
Background technique
Our living environment is a three-dimensional even world of multidimensional, currently, the object that we are touched is mostly
Three-dimension object has its unique inner molecular structure and outer shape structure, to give a kind of space multistory sense of people.In recent years
Come, with the development of modeling technique and scanning technique, applying for threedimensional model all achieves good achievement in every field, packet
Include the various aspects such as 3d gaming, Design of Industrial Product, Building Design, medical diagnosis, virtual reality and video display animation.With
The extension of the application field range of threedimensional model, every technical need of threedimensional model also becomes increasing, including three-dimensional mould
The various aspects such as type identification, reconstruct, matching, segmentation, classification, retrieval.In each application scenarios, category of model link is for it
It is most important for his link, such as: during classification can to model carry out feature on understanding and can by feature into
Matching of the row storage to the later period；Classification results can be used for retrieving and the study etc. of unsupervised segmentation algorithm.Therefore, three-dimensional mould
The classification method of type is one of the hot spot in three-dimensional model studying.
In recent years, scholars propose the feature that many feature descriptors are used to describe threedimensional model, such as shape diameter letter
Number (Shape Diameter Function, SDF) and Gaussian curvature (Gaussian Curvature, GC), average geodesic distance
(Average Geodesic Distance, AGD) and distribution of shapes (D2) etc..Current existing feature descriptor can be by
Be summarized as two classes: one kind is complete acute feature descriptor, and this feature descriptor describes the geometrical characteristic of entire threedimensional model；One
Kind is local feature description's symbol, and this feature descriptor can only describe the geometrical characteristic of the part of threedimensional model.Due to three-dimensional mould
The geometric complexity and space multistory complexity of type, can be very good to reflect all classes there is presently no a kind of feature descriptor
The different characteristic of the threedimensional model of type, for this case, scholars usually extract three using various features descriptor simultaneously
The feature of dimension module, can portray the geometrical characteristic of threedimensional model from multiple angles in this way, and multiple features combine to obtain
More accurately classification results.After carrying out feature extraction to threedimensional model using feature descriptor, need through clustering algorithm pair
Feature vector carries out clustering, to distinguish generic.In recent years, scholars propose many clustering algorithms, such as K-
Means (K mean value) cluster, mean shift clustering, density clustering method (DBSCAN), cohesion level are poly-, figure group examines
It surveys (Graph Community Detection), clustered etc. with the greatest hope (EM) of gauss hybrid models (GMM).
The Gaussian curvature (Gaussian Curvature, GC) of certain point is this principal curvatures K1 and K2 on threedimensional model
Product, it is the inherent measurement of curvature, its value only depends on how the distance on curved surface measures；Shape diameter function (Shape
Diameter Function, SDF) it is defined in a real-valued function on each point in threedimensional model surface, it expresses grid
The measurement method of adjacent spots object volume diameter on surface, based on the shape function based on volume, largely
The similar portion maintenance similar value for keeping independence and different objects can be changed to the posture of same object.Thus it summarizes
It sees, GC features the Gaussian curvature on each vertex on threedimensional model；SDF features the diameter length of threedimensional model different location.
In recent years, many scholars both domestic and external conduct in-depth research threedimensional model classification problem, and propose one
The effective sorting algorithm of series, such as: the threedimensional model sorting algorithm based on semantic information and shared conceptual model, the calculation
Method first carries out label to each threedimensional model using K-means method in shape indexing space, is then based on semantic tagger pair
Threedimensional model is classified；Threedimensional model is solved using radial basis function (Radial basis function) neural network
Classification, the algorithm are weighted and averaged by the classification information that the range information to threedimensional model is exported with neural network, finally
Obtain the similitude between threedimensional model；Threedimensional model point is solved based on the semi-supervised learning method of transductive SVM
Class problem etc..
Currently based on threedimensional model feature descriptor classification method realization process mainly there are several types of:
1, by the threedimensional model sorting algorithm of local rarefaction representation:
This algorithm is classified for the threedimensional model of unknown classification information, and main flow is: first with one kind
Feature is extracted to threedimensional model based on feature descriptor-shape diameter function (Shape Diameter Function, SDF)
Vector, and using the threedimensional model of unknown classification information as test model, it is found in the three-dimensional modeling data storehouse of known classification
The K model most like with test model；Then rarefaction representation classification (sparse is utilized in this K model
Representation classifier, SRC) method identifies test model, finally determine test model in three-dimensional mould
Classification information in type library.
2, by the threedimensional model sorting algorithm of core principle component analysis:
This is a kind of based on core principle component analysis (Kernel-Principal Components A-nalysis, K-PCA)
Threedimensional model sorting algorithm.This algorithm selected shape diameter function first (Shape Diameter Function, SDF)
The feature vector of threedimensional model is extracted as feature descriptor, and original feature vector is then mapped to higher-dimension using kernel function
It in space and carries out PCA on this space and obtains new feature vector；Finally using KNN algorithm and calculate Unknown Model with it is known
L2 norm between K model of classification is with the classification of implementation model, so that it is determined that the classification of Unknown Model.
Both the above method is compared, and second method has done further processing to feature vector, by feature vector
Core principle component analysis operation is carried out；Two methods are different to the feature vector Processing Algorithm finally obtained, therefore for feature
The selection of descriptor and the mode that characteristic vector data is handled for threedimensional model classification results for be our needs
The emphasis of research.
Although more to the method comparison of threedimensional model classification in the prior art, classification results can also be obtained in its application field
To good application, but the research of threedimensional model classification is underway always, it is most important that for the demand of classification results
It is more and more rigorous, the size control of error is become smaller, the scope control for the property of would be compatible with becomes larger.It is summarized from most of research and analyse
Out: on the one hand, the classification of threedimensional model is mainly by its of certain characteristic value or semantic information or a kind of feature descriptor
It is middle a kind of as basic data, but one of basic data can only depict a kind of geometrical characteristic of threedimensional model, for
It is relatively simple for the threedimensional model of geometry complexity, a variety of threedimensional models cannot be suitable for well；On the other hand, to mentioning
The characteristic value taken carries out certain algorithm and analyzes to obtain new feature vector, and uses algorithm by new feature vector and known classification
As a result model data carries out calculating ratio pair so that it is determined which classification it belongs to, for traditional PCA calculation in 3 d model library
Although method can be used for dimensionality reduction, but be easy to be influenced by the larger noise of individual intensity, this process is comparatively not rigorous enough,
There is certain limitation to the requirement of threedimensional model in use.The disadvantage present in these threedimensional model assorting processes is main
With the presence of basic data is relatively simple and alignment algorithm limitation process is not rigorous enough, thus will lead to sorting algorithm compatibility compared with
The problems such as low and classification results error is larger.
Summary of the invention
Aiming at the shortcomings existing in the above problems, the present invention provides a kind of three-dimensional mould based on various features descriptor
Type classification method.
The invention discloses a kind of threedimensional model classification methods based on various features descriptor, comprising:
Various features descriptor is selected to extract the feature vector of threedimensional model, the feature descriptor includes Gauss song
Rate, shape diameter function and scale invariant feature conversion；
Denoising dimension-reduction treatment is carried out to described eigenvector using Robust Principal Component Analysis；
It is clustered using density-based algorithms, obtains threedimensional model classification results.
As a further improvement of the present invention, the Gaussian curvature collection that threedimensional model is obtained using the library PCL, can by curvature value
Position and its distribution of the biggish point of curvature are verified on threedimensional model depending on changing.
As a further improvement of the present invention, the calculation method of the shape diameter function are as follows:
Any point in three-dimensional point cloud is taken, is constructed using the negative normal direction of point as the cone of center axis, then calculating has
Imitate the weighted average length of ray segment, the shape diameter functional value as the point.
As a further improvement of the present invention, described to be clustered using density-based algorithms, obtain three-dimensional mould
Type classification results；Include:
M threedimensional model sample is given, sample set D=(x is set1,x2,...,xm), Neighbourhood parameter (∈, MinPts), sample
This distance metric mode；Calculate each feature vector xiLocal density and lowest distance value:
xiLocal density beWherein dijIllustrate xiAnd xjBetween Euclidean distance；dkFor truncation away from
From ρiThose and x are measurediThe distance between in dkWithin point number；
xiThe minimum value of the distance between those density ratios its big feature vector isThe parameter is used for
Those local densities are greater than xiFeature vector in find out and xiBetween the shortest distance；
This algorithm passes through ψi=ρi′γiTo determine the cluster centre of each threedimensional model type, ψiIt is maximum before several
Feature vector is considered as cluster centre；Each cluster centre is formed into a set F=(ψ1,ψ2,...,ψm), then by it
The ρ and ψ that it is not belonging to the corresponding feature vector of threedimensional model of cluster centre traverse set F according to algorithm, for each spy
Levy vector xi, find that feature vector x nearest in all density ratio others feature vectorsi', then xiClassification category
In xi' corresponding ψiClassification.
Compared with prior art, the invention has the benefit that
The present invention, which controls basic data using various features descriptor, can avoid unicity, to the obtained feature of extraction to
Amount denoising dimensionality reduction is simultaneously clustered using density-based algorithms, can get the result of more accurate classification.
Detailed description of the invention
Fig. 1 is the process of the threedimensional model classification method based on various features descriptor disclosed in an embodiment of the present invention
Figure；
Fig. 2 is that threedimensional model Gaussian curvature disclosed in an embodiment of the present invention visualizes schematic diagram；
Fig. 3 is that threedimensional model disclosed in an embodiment of the present invention visualizes schematic diagram；
Fig. 4 is the calculation method exemplary diagram of SDF value disclosed in an embodiment of the present invention.
Specific embodiment
In order to make the object, technical scheme and advantages of the embodiment of the invention clearer, below in conjunction with the embodiment of the present invention
In attached drawing, technical scheme in the embodiment of the invention is clearly and completely described, it is clear that described embodiment is
A part of the embodiments of the present invention, instead of all the embodiments.Based on the embodiments of the present invention, ordinary skill people
Member's every other embodiment obtained without making creative work, shall fall within the protection scope of the present invention.
Threedimensional model has its unique inner molecular structure and outer shape structure, as the application of threedimensional model is more next
More extensive, the research of the classification method of threedimensional model is underway always, but there are also permitted for the classification method of threedimensional model
Need perfect place, for example whether the process classified is rigorous, whether the control of classification results error size, classification results are simultaneous more
Hold the threedimensional model of other field.Each feature descriptor can be very good to portray the local feature of threedimensional model, therefore, this hair
It is bright to propose a kind of threedimensional model classification method based on various features descriptor, basic number is controlled using various features descriptor
According to avoiding feature unicity, and the feature vector extracted by feature descriptor is subjected to denoising dimension-reduction treatment obtains new feature
Then vector carries out the result that rigorous density-based algorithms analysis obtains more accurate classification.
The present invention is described in further detail with reference to the accompanying drawing:
As shown in Figure 1, the present invention provides a kind of threedimensional model classification method based on various features descriptor, comprising:
S1, various features descriptor is selected to extract the feature vector of threedimensional model；
It is specific:
Feature descriptor is widely used in threedimensional model high level analysis field, and the geometry that it can portray threedimensional model is special
It seeks peace topological structure.For threedimensional model, a kind of feature descriptor cannot come out its all feature extraction, also without
Method is suitable for all threedimensional models, therefore algorithm of the invention needs the multiple feature descriptors of simultaneous selection to extract threedimensional model
Feature: Gaussian curvature (Gaussian Curvature, GC), shape diameter function (Shape Diameter Function,
SDF) and scale invariant feature converts (Scale-invariant feature transform, SIFT).Wherein:
Gaussian curvature (Gaussian Curvature, GC): the Gaussian curvature that certain is put on threedimensional model curved surface, the i.e. point two
The product of a principal curvatures.Vertex on curved surface is mapped to the centre of sphere of unit ball, the endpoint of normal is mapped on spherical surface, i.e., will
Point on curved surface establishes a kind of corresponding with the point on spherical surface, is called spheric representation of a surface, is also Gauss Map.Gaussian curvature
Geometric meaning, i.e., area/curved surface local area limit on spherical surface, it can be seen that Gaussian curvature reflects curved surface really
The bending degree of part.The library PCL can be used in experiment to obtain the Gaussian curvature collection of threedimensional model, curvature value visualization is existed
Position and its distribution of the biggish point of curvature can be verified on threedimensional model, comparison is as shown in Figure 2,3.
Shape diameter function (Shape Diameter Function, SDF): enabling M for two-dimentional flow pattern triangle gridding surface,
A scalar function fv:M → R is defined as the neighborhood of a point diameter in surface mesh, this function is referred to as shape diameter function.
The present invention uses a kind of simple method for calculating shape diameter function, core concept are as follows: first in taking three-dimensional point cloud
Any point, the normal direction of point are obtained according to steady normal estimation method, are constructed using the negative normal direction of point as the circle of center axis
Then cone calculates the weighted average length of effective rays section, the SDF value as the point.The circular of SDF value is such as
Under, as shown in Figure 4: from any point p on 3 d discrete point cloud modeliIt sets out, then with point piNegative normal direction be center axis structure
A cone C is made, the subtended angle of cone is θ, takes the intracorporal point q of the circular conej, calculate point piWith point qjRay segment rjLength
Degree is lj, count ljMean valueAnd variance Then it chooses and is located at penetrating within the scope of 1 variance of mean value
Line segment, i.e., all satisfactionsRay segment ri, by riAscending sort is carried out, k ray segment lj is as effectively before selecting
Ray segment.Calculating chooses the weighted average dependent variable q of ray segment to indicate, i.e.,As SDF feature in point
piOn value, take corresponding ray segment lkTo weight wi of the inverse as the ray segment of angle between circular cone central axes.
Scale invariant feature conversion (Scale-invariant feature transform, SIFT) is a kind of computer view
The algorithm of feel is used to the locality characteristic of descriptive model, it finds extreme point in space scale, and can extract its position, ruler
Degree, rotational invariants.SIFT algorithm can solve to a certain extent the rotation of target, scaling, translation (RST), image it is affine/throw
Shadow converts the problems such as (viewpoint viewpoint), target occlusion (occlusion), noise.The main feature of SIFT algorithm has:
SIFT feature is the local feature of image, is maintained the invariance to rotation, scaling, brightness change, to visual angle change, affine
Transformation, noise also keep a degree of stability；Unique (Distinctiveness) is good, informative, is suitable for
It is fast and accurately matched in magnanimity property data base.
S2, denoising dimension-reduction treatment is carried out to feature vector using Robust Principal Component Analysis；
It is specific:
It is also similar in the feature vector that appropriate feature descriptor extracts for similar threedimensional model.That
Matrix composed by the corresponding feature vector of multiple threedimensional models should be close to low-rank, but in the structure of threedimensional model
During building due to various reasons, may be such that data, there are some noises, and the spy obtained by various features descriptor
The dimension for levying vector is higher, and dimension is excessively high to will cause data redundancy problem, therefore we need the spy of feature descriptor extraction
Sign vector carries out denoising dimension-reduction treatment.
It is obtained according to the analysis of first part, uses traditional principal component point in major part threedimensional model assorting process at present
It analyses (principal component analysis, PCA) algorithm and dimension-reduction treatment is carried out to the feature vector of threedimensional model.Tradition
Principal component analysis (principal component analysis, PCA) can effectively find out most important member in data
Element and structure remove noise and redundancy, original complex data can be carried out dimensionality reduction.Simplest principal component analytical method is exactly
PCA, from the point of view of linear algebra, the target of PCA is exactly the new data space for going to redescribe using another group of base,
New base is organized by this, can disclose the relationship between original data, i.e. this dimension is most important " pivot ".The target of PCA
It exactly finds such " pivot ", removes the interference of redundancy and noise to the greatest extent.
As classical PCA, Robust PCA (Robust Principal Component Analysis) is substantially also to find data in lower dimensional space
On best projection problem.When observe data it is larger when, PCA can not provide it is ideal as a result, and Robust PCA can from compared with
The data of substantially low-rank are recovered in the observation data of big and sparse noise pollution.What Robust PCA considered is such one
A problem: general data matrix D includes structural information, also includes noise.It can be so two squares by this matrix decomposition
Battle array is added: D=A+E, A are low-rank (being linearly related between having certain structural information to cause each row or column due to inside), E
It is sparse (being then sparse containing noise), then Robust PCA can be write as optimization problem below:
Since there are non-convex and Non-smooth surface characteristics in optimization for rank and L0 norm, so generally this np problem is converted
At one loose convex optimization problem of solution:
S3, it is clustered using density-based algorithms, obtains threedimensional model classification results；
It is specific:
Density-based algorithms (Density-Based Spatial Clustering of Application
With Noise, DBSCAN), be a kind of very typical density clustering algorithm and K-Means, BIRCH these be generally only applicable to
The clustering method of convex sample set is compared, and DBSCAN both can be adapted for convex sample set, is readily applicable to non-convex sample set.
DBSCAN is a kind of density-based algorithms, and this kind of density clustering algorithm commonly assumes that classification can be by sample distribution
Tightness degree determines.Same category of sample, they be characterized in it is extremely similar, feature vector be also it is very similar, in this way
It may determine that there are the samples of same classification within the scope of predetermined radius around this classification sample.
Data point is divided into three classes in DBSCAN algorithm:
Core point: it is included in radius r and has more than MinPts number point of destination；
Boundary point: the quantity put in radius r is less than MinPts, but falls in core neighborhood of a point；
Noise point: neither core point is also not the point of boundary point；
Herein there are two amount, one is radius r, the other is specified number MinPts.
Specific step is as follows for algorithm:
1. determining radius r and MinPts. first since a not visited arbitrary number strong point, it is with this point
Whether the quantity at center, point of the r to include in the circle of radius is greater than or equal to MinPts, then changes if it is greater than or equal to MinPts
Point is marked as cPoint, on the contrary then can be marked as nPoint.
2. the step of repeating 1, if a nPoint is present in the circle that some cPoint is radius, this point is marked
It is denoted as marginal point, otherwise is still nPoint.Repeat step 1, it is known that all points are all accessed.
For threedimensional model classification the present invention in the following method:
M threedimensional model sample is given, sample set D=(x is set1,x2,...,xm), Neighbourhood parameter (∈, MinPts), sample
This distance metric mode；Calculate each feature vector xiTwo parameters, i.e. local density and lowest distance value:
1、xiLocal density beWherein dijIllustrate xiAnd xjBetween Euclidean distance；dkFor truncation
Distance, ρiThose and x are measurediThe distance between in dkWithin point number；
2、xiThe minimum value of the distance between those density ratios its big feature vector isThe parameter is used for
It is greater than x in those local densitiesiFeature vector in find out and xiBetween the shortest distance；
This algorithm passes through ψi=ρi′γiTo determine the cluster centre of each threedimensional model type, ψiIt is maximum before several
Feature vector is considered as cluster centre；Each cluster centre is formed into a set F=(ψ1,ψ2,...,ψm), then by it
The ρ and ψ that it is not belonging to the corresponding feature vector of threedimensional model of cluster centre traverse set F according to algorithm, for each spy
Levy vector xi, find that feature vector x nearest in all density ratio others feature vectorsi', then xiClassification category
In xi' corresponding ψiClassification.
These are only the preferred embodiment of the present invention, is not intended to restrict the invention, for those skilled in the art
For member, the invention may be variously modified and varied.All within the spirits and principles of the present invention, it is made it is any modification,
Equivalent replacement, improvement etc., should all be included in the protection scope of the present invention.
Claims (4)
1. a kind of threedimensional model classification method based on various features descriptor characterized by comprising
Various features descriptor is selected to extract the feature vector of threedimensional model, the feature descriptor includes Gaussian curvature, shape
Shape diameter function and scale invariant feature conversion；
Denoising dimension-reduction treatment is carried out to described eigenvector using Robust Principal Component Analysis；
It is clustered using density-based algorithms, obtains threedimensional model classification results.
2. as described in claim 1 based on the threedimensional model classification method of various features descriptor, which is characterized in that use
The library PCL obtains the Gaussian curvature collection of threedimensional model, and curvature value visualization verified the biggish point of curvature on threedimensional model
Position and its distribution.
3. as described in claim 1 based on the threedimensional model classification method of various features descriptor, which is characterized in that the shape
The calculation method of shape diameter function are as follows:
Any point in three-dimensional point cloud is taken, is constructed using the negative normal direction of point as the cone of center axis, is then calculated and effectively penetrate
The weighted average length of line segment, the shape diameter functional value as the point.
4. as described in claim 1 based on the threedimensional model classification method of various features descriptor, which is characterized in that described to make
It is clustered with density-based algorithms, obtains threedimensional model classification results；Include:
M threedimensional model sample is given, sample set D=(x is set1,x2,...,xm), Neighbourhood parameter (∈, MinPts), sample away from
From metric form；Calculate each feature vector xiLocal density and lowest distance value:
xiLocal density beWherein dijIllustrate xiAnd xjBetween Euclidean distance；dkFor distance, ρ is truncatedi
Those and x are measurediThe distance between in dkWithin point number；
xiThe minimum value of the distance between those density ratios its big feature vector isThe parameter is used at those
Local density is greater than xiFeature vector in find out and xiBetween the shortest distance；
This algorithm passes through ψi=ρi×γiTo determine the cluster centre of each threedimensional model type, ψiSeveral features before maximum
Vector is considered as cluster centre；Each cluster centre is formed into a set F=(ψ1,ψ2,...,ψm), then by it is other not
The ρ and ψ for belonging to the corresponding feature vector of threedimensional model of cluster centre traverse set F according to algorithm, for each feature to
Measure xi, find that feature vector x nearest in all density ratio others feature vectorsi', then xiClassification belong to xi’
Corresponding ψiClassification.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910123038.0A CN109829505A (en) | 2019-02-15 | 2019-02-15 | A kind of threedimensional model classification method based on various features descriptor |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910123038.0A CN109829505A (en) | 2019-02-15 | 2019-02-15 | A kind of threedimensional model classification method based on various features descriptor |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109829505A true CN109829505A (en) | 2019-05-31 |
Family
ID=66862213
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910123038.0A Pending CN109829505A (en) | 2019-02-15 | 2019-02-15 | A kind of threedimensional model classification method based on various features descriptor |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109829505A (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110619364A (en) * | 2019-09-18 | 2019-12-27 | 哈尔滨理工大学 | Wavelet neural network three-dimensional model classification method based on cloud model |
| CN110781918A (en) * | 2019-09-23 | 2020-02-11 | 辽宁师范大学 | Non-rigid three-dimensional model classification algorithm for self-adaptive sparse coding fusion |
| CN112258650A (en) * | 2020-09-21 | 2021-01-22 | 北京科技大学 | Paste filling progress real-time measuring and visualizing method and system |
| CN113297879A (en) * | 2020-02-23 | 2021-08-24 | 深圳中科飞测科技股份有限公司 | Acquisition method of measurement model group, measurement method and related equipment |
| CN113705336A (en) * | 2021-07-15 | 2021-11-26 | 南京林业大学 | Flexible cutting smoke robust feature extraction method |
| CN116434220A (en) * | 2023-04-24 | 2023-07-14 | 济南大学 | Three-dimensional object classification method and system based on descriptor and AdaBoost algorithm |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105354593A (en) * | 2015-10-22 | 2016-02-24 | 南京大学 | NMF (Non-negative Matrix Factorization)-based three-dimensional model classification method |
-
2019
- 2019-02-15 CN CN201910123038.0A patent/CN109829505A/en active Pending
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105354593A (en) * | 2015-10-22 | 2016-02-24 | 南京大学 | NMF (Non-negative Matrix Factorization)-based three-dimensional model classification method |
Non-Patent Citations (3)
| Title |
|---|
| SHAPIRA L ET AL.: "Consistent mesh partitioning and skeletonisation using the shape diameter function", 《THE VISUAL COMPUTER》 * |
| 王鹏飞: "一种三维模型识别算法研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
| 祁成武: "三维模型分割与分类算法研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110619364A (en) * | 2019-09-18 | 2019-12-27 | 哈尔滨理工大学 | Wavelet neural network three-dimensional model classification method based on cloud model |
| CN110781918A (en) * | 2019-09-23 | 2020-02-11 | 辽宁师范大学 | Non-rigid three-dimensional model classification algorithm for self-adaptive sparse coding fusion |
| CN110781918B (en) * | 2019-09-23 | 2023-06-23 | 辽宁师范大学 | Non-rigid three-dimensional model classification algorithm based on self-adaptive sparse coding fusion |
| CN113297879A (en) * | 2020-02-23 | 2021-08-24 | 深圳中科飞测科技股份有限公司 | Acquisition method of measurement model group, measurement method and related equipment |
| CN112258650A (en) * | 2020-09-21 | 2021-01-22 | 北京科技大学 | Paste filling progress real-time measuring and visualizing method and system |
| CN112258650B (en) * | 2020-09-21 | 2024-03-19 | 北京科技大学 | Paste filling progress real-time measurement and visualization method and system |
| CN113705336A (en) * | 2021-07-15 | 2021-11-26 | 南京林业大学 | Flexible cutting smoke robust feature extraction method |
| CN113705336B (en) * | 2021-07-15 | 2024-03-19 | 南京林业大学 | Flexible cutting smoke robust feature extraction method |
| CN116434220A (en) * | 2023-04-24 | 2023-07-14 | 济南大学 | Three-dimensional object classification method and system based on descriptor and AdaBoost algorithm |
| CN116434220B (en) * | 2023-04-24 | 2024-02-27 | 济南大学 | Three-dimensional object classification method and system based on descriptor and AdaBoost algorithm |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109829505A (en) | A kind of threedimensional model classification method based on various features descriptor | |
| Gorelick et al. | Shape representation and classification using the poisson equation | |
| CN107742102B (en) | Gesture recognition method based on depth sensor | |
| Li et al. | SHREC’14 track: Extended large scale sketch-based 3D shape retrieval | |
| Rusu et al. | Learning informative point classes for the acquisition of object model maps | |
| Wang et al. | A perception-driven approach to supervised dimensionality reduction for visualization | |
| CN104392253B (en) | Interactive classification labeling method for sketch data set | |
| Kurnianggoro et al. | A survey of 2D shape representation: Methods, evaluations, and future research directions | |
| CN103295025B (en) | A kind of automatic selecting method of three-dimensional model optimal view | |
| Wenjing et al. | Research on areal feature matching algorithm based on spatial similarity | |
| CN109447100A (en) | A kind of three-dimensional point cloud recognition methods based on the detection of B-spline surface similitude | |
| CN108388902B (en) | Composite 3D descriptor construction method combining global framework point and local SHOT characteristics | |
| Zhao et al. | Character‐object interaction retrieval using the interaction bisector surface | |
| CN111460193B (en) | Three-dimensional model classification method based on multi-mode information fusion | |
| CN109271441A (en) | A kind of visualization clustering method of high dimensional data and system | |
| Hu et al. | MAT-Net: Medial Axis Transform Network for 3D Object Recognition. | |
| Çuğu et al. | Treelogy: A novel tree classifier utilizing deep and hand-crafted representations | |
| Zafari et al. | Segmentation of partially overlapping convex objects using branch and bound algorithm | |
| Bai | Scene categorization through using objects represented by deep features | |
| Chaudhuri et al. | Exploring flow fields using space-filling analysis of streamlines | |
| CN105975906A (en) | PCA static gesture recognition method based on area characteristic | |
| Lee | Symmetry-driven shape description for image retrieval | |
| Pratikakis et al. | Partial 3d object retrieval combining local shape descriptors with global fisher vectors | |
| Shi et al. | Integral curve clustering and simplification for flow visualization: A comparative evaluation | |
| CN110334704B (en) | Three-dimensional model interest point extraction method and system based on layered learning |
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: 20190531 |