CN106157375B - A kind of threedimensional model component categories automatic marking method - Google Patents
A kind of threedimensional model component categories automatic marking method Download PDFInfo
- Publication number
- CN106157375B CN106157375B CN201610530040.6A CN201610530040A CN106157375B CN 106157375 B CN106157375 B CN 106157375B CN 201610530040 A CN201610530040 A CN 201610530040A CN 106157375 B CN106157375 B CN 106157375B
- Authority
- CN
- China
- Prior art keywords
- dough sheet
- model
- feature
- mark
- grid edge
- 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.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T19/00—Manipulating 3D models or images for computer graphics
- G06T19/20—Editing of 3D images, e.g. changing shapes or colours, aligning objects or positioning parts
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- 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
- G06T2207/20072—Graph-based image processing
-
- 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
-
- 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
- G06T2207/20084—Artificial neural networks [ANN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2219/00—Indexing scheme for manipulating 3D models or images for computer graphics
- G06T2219/004—Annotating, labelling
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Software Systems (AREA)
- Computer Graphics (AREA)
- Architecture (AREA)
- Computer Hardware Design (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of threedimensional model component categories automatic marking methods, include the following steps：Training process marks training set according to threedimensional model and is quickly trained, the quick marking model for training the dough sheet obtained for unknown threedimensional model and Grid Edge to be classified and marked.The quick marking model of dough sheet and Grid Edge that annotation process is then obtained using training classifies to the dough sheet and Grid Edge of object module, obtain dough sheet and the distribution of Grid Edge class probability, construct graph model, optimization is cut by multi-tag figure and is split the smooth of boundary and optimization, to realize the fast automatic mark to target three-dimensional component categories.
Description
Technical field
The present invention relates to a kind of processing methods of shape analysis, belong to computer graphics techniques field, specifically
A kind of threedimensional model component categories automatic marking method.
Background technique
It is the basis of shape understanding and high-level geometric manipulations to the understanding that threedimensional model component is constituted.In recent years, it studies
The machine learning techniques of mature are gradually introduced into 3D shape analysis and research by personnel, the geometrical model of data-driven
Joint dividing method just more and more attention has been paid to.It, can be in input using " the information transmitting " of data-driven this key concept
A kind of correlation is established between the sample shape for preassigning or calculating and the target shape for needing to analyze and handle, and can will be felt
The information of interest is transmitted to target shape from sample shape.Therefore, the shape segmentations of data-driven can produce having for consistency
The component of semantic tagger, and can effectively support the subsequent geometry such as the synthesis of threedimensional model, example modeling and textures synthesis
Processing task.
Based on whether there are available label for labelling data as input information in study, 3D shape can be divided and dividing
For：Supervised segmentation, non-formaldehyde finishing and semi-supervised segmentation.In general, supervised learning method passes through given mark number
According to tend to output with it is intended that be close as a result, to obtain higher segmentation precision.There are supervision shape segmentations usual
Form turns to a classification problem.Such as document 1：KALOGERAKIS E.,HERTZMANN A.,SINGH K.:Learning 3d
mesh segmentation and labeling.ACM Trans.Graph.29(2010),102:1–102:12. segmentation is asked
Topic form turns to the optimization problem of condition random field CRF.Its condition random field CRF model, it is consistent with label comprising evaluation dough sheet
Property unitary item and adjacent dough sheet tag compliance binary item, unitary item and binary Xiang Junke are logical according to markup information
JointBoost classifier is crossed to learn to obtain.Document 2：GUO K.,Zou D.,and Chen X.3D Mesh Labeling
via Deep Convolutional Neural Networks[J].ACM Transactions on Graphics,2015,
35(1):3:1-3:12. by deep neural network from the acquistion of data middle school to a kind of effective and robust shape representation, into
One step improves segmentation and mark performance.However, the training process of the above method be it is very time-consuming, such as only 6 shapes are carried out
When training, training process just needs to expend several hours.In order to improve training effectiveness, document 3：XIE Z.,XU K.,LIU L.,
XIONG Y.:3d shape segmentation and labeling via extreme learning
33,5 (2014) of machine.Computer Graphics Forum then carries out the classification of dough sheet, training based on ELM classifier
Process is greatly speeded up, averagely only need to be less than 1 minutes when 6 shapes of training.However, this method is only only accounted for according to instruction
Practice the dough sheet markup information in data to generate a dough sheet classifier, there is no believe using the boundary gone out given in labeled data
Breath, but rough boundary position is obtained by carrying out over-segmentation to target shape, then optimize, final segmentation result
Quality depend heavilys on the result of over-segmentation.Method disclosed by the invention then passes through the activation primitive of neuron in modification ELM,
Classifying quality and the classification time of ELM are further adjusted, and the boundary information in labeled data is also made full use of.According to
Dough sheet classification annotation probability distribution and the distribution of Grid Edge class probability, construct graph model, cut optimization method essence by multi-tag figure
It adjusts and obtains more smooth and accurate segmentation result.
Summary of the invention
Goal of the invention：It is a kind of quick the technical problem to be solved by the present invention is in view of the deficiencies of the prior art, provide
Threedimensional model component categories automatic marking method carries out rapidly automatic segmentation and mark to threedimensional model for supporting.
In order to solve the above-mentioned technical problem, the invention discloses a kind of threedimensional model component categories automatic marking method, packets
Include following steps：
Step 1, training threedimensional model marks training set, and the threedimensional model in training set is with component in class model and model
Standard mark is provided, wherein markup information includes the classification mark that each patch grids is subordinated to component parts in threedimensional model
And every Grid Edge is subordinated to the classification mark of boundary edge, is quickly trained, is obtained by marking training set to threedimensional model
The quick dough sheet marking model and quick Grid Edge marking model with mark must be used to be split target three-dimensional, and obtained
Threedimensional model dough sheet classification annotation probability distribution and Grid Edge classification annotation probability distribution are obtained, and study obtains figure and cuts the flat of model
Sliding item weight；
Step 2, label target threedimensional model：The quick dough sheet marking model obtained using step 1 and quick Grid Edge mark
Injection molding type is split and marks to target three-dimensional respectively, and each dough sheet for obtaining target three-dimensional is subordinated to constituting portion
The classification mark of part and every Grid Edge are subordinated to the classification mark of boundary edge, construct graph model, are cut by multi-tag figure excellent
Change obtains smooth segmentation and annotation results.
Step 1 includes the following steps：
Step 1-1 pre-processes threedimensional models all in training set, extract each patch grids dough sheet feature and
The side feature of every Grid Edge；
Step 1-2, all dough sheet composing training patch grids collection of threedimensional model in training set, using the improved limit
Habit machine method (ELM, extreme learning machine), to the dough sheet for each patch grids that training patch grids is concentrated
Feature and its standard mark are quickly trained, and quick dough sheet marking model is obtained；
Step 1-3, all Grid Edge composing training Grid Edge collection of threedimensional model in training set, using the improved limit
The side feature and its standard mark for every Grid Edge that habit machine method concentrates training Grid Edge are quickly trained, and are obtained quick
Grid Edge marking model.
Step 1-4 obtains threedimensional model dough sheet classification annotation probability distribution and Grid Edge classification annotation probability distribution, and learns
The smooth item weight of model is cut in acquistion to figure.
Carrying out pretreatment to threedimensional models all in training set described in step 1-1 includes model normalized and feature
It extracts, wherein model normalized includes the following steps：
The mass center of threedimensional model in training set is moved to coordinate origin by step 1-1-1, and the mass center of threedimensional model passes through
The area weighted average on all vertex obtains on computation model；
Step 1-1-2 calculates each dough sheet center of threedimensional model in training set to the Euclidean distance of its mass center, takes all
The intermediate value of Euclidean distance completes instruction by the coordinate of each point on the threedimensional model in training set divided by the specification item as specification item
Practice the normalized for the threedimensional model concentrated；
Feature extraction includes the following steps：
Step 1-1-3, it is special by curvature feature, PCA by the dough sheet feature of each patch grids of threedimensional model in training set
Sign, shape characteristics of diameters, average geodesic distance feature and Shape context feature cascade to form a feature vector；
Step 1-1-4, by the side feature of every Grid Edge of threedimensional model in training set, by while dihedral angle feature, while it is adjacent
The dihedral angle feature on domain vertex, two adjacent dough sheets of the curvature difference and derivative feature on side two sides neighborhood vertex, shared Grid Edge
The shape image difference sign cascade of shape diameter difference feature, two adjacent dough sheets for sharing side forms a feature vector.
Step 1-2 includes the following steps：
Training patch grids is concentrated the dough sheet feature of each dough sheet as Single hidden layer feedforward neural networks by step 1-2-1
Input unit value inputted；
Step 1-2-2, modeling Single hidden layer feedforward neural networks are：
Wherein, j=1,2 ... N, N are the quantity that training set patch grids concentrates dough sheet, xjIt is special for the corresponding dough sheet of dough sheet j
Vector is levied,For the number of hidden layer neuron；wi=(wi1,wi2,…,wiN)TIndicate that hidden layer neuron and input unit connect
The weight vector connect may be configured as the random number between (- 1,1)；For offset vector, (0,1) may be configured as
Between random number；G (y) is activation primitive；ljFor according to the reality output of dough sheet j in the resulting training set of model, βkFor output
The output weight that unit is connect with hidden layer, k=1,2 ... m, m are the dimension of output layer, that is, the number of tags classified；
Step 1-2-3 selects neuron activation function ReLu (Rectified Linear Units) function of approximate biology
As activation primitive, it can have preferably unilateral inhibition compared to traditional sigmoid activation primitive, obtain relatively broad
Excited boundary obtains sparse activity, it is hereby achieved that trained speed and more stable training precision faster.ReLu
Activation primitive is defined as：G (y)=max (0, y).
Step 1-2-4, when approaching the model with zero error,I.e.：
Wherein, tjFor the mark of standard corresponding to dough sheet each in training set, g is the neuron activation selected in step 1-2-3
Function, the weight matrix between hidden layer and output layer are represented by：
Wherein,Indicate the weight vector of output unit and r-th of hidden layer neuron,
Step 1-2-5 is approached according to zero error is carried out to model in step 1-2-4, is provided matrix and be expressed as H β=T,
In, T=(t1 t2 … tj), tjFor the mark of standard corresponding to dough sheet j in training set；G is the nerve selected in step 1-2-3
Activation primitive.Pass through solutionObtain output weight.For the Moore-Penrose of matrix
Generalized inverse.So far the quick marking model of dough sheet has been obtained.
Using the side feature and its standard mark of the ELM every Grid Edge concentrated to training Grid Edge described in step 1-3
It is quickly trained, obtains quick Grid Edge marking model, include the following steps：
Training Grid Edge is concentrated the side feature of each edge as the input of Single hidden layer feedforward neural networks by step 1-3-1
Cell value is inputted.
Step 1-3-2, modeling Single hidden layer feedforward neural networks are：
Wherein, ej=1,2 ... Ne, NeThe quantity of Grid Edge, x are concentrated for training Grid EdgeejFor the corresponding side Grid Edge ej
Feature,For the number of hidden layer neuron, wei=(wei1,wei2,…,weiN)TIndicate that hidden layer neuron and input unit connect
The weight vector connect may be configured as the random number between (- 1,1)；For offset vector, may be configured as (0,
1) random number between；G (y) is activation primitive, lejFor according to the reality output of the resulting Grid Edge ej of model, βekFor output
The output weight that unit is connect with hidden layer, ek=1,2 ... me, meFor the dimension of output layer, that is, the number of tags classified；
Step 1-3-3 selects neuron activation function ReLu (Rectified Linear Units) function of approximate biology
As activation primitive, it can have preferably unilateral inhibition compared to traditional sigmoid activation primitive, obtain relatively broad
Excited boundary obtains sparse activity, it is hereby achieved that trained speed and more stable training precision faster.ReLu
Activation primitive is defined as：G (y)=max (0, y).
Step 1-3-4, when approaching the model with zero error,I.e.：
Wherein, tej∈ { 1,2 } is the mark of standard corresponding to Grid Edge ej in training set, respectively corresponds non-boundary edge and side
Boundary side, g are the neuron activation function selected in step 1-3-3, and the weight matrix between hidden layer and output layer is represented by：
Wherein,Indicate the weight vector of output unit and first of hidden layer neuron,For the quantity of hidden layer neuron, meFor output layer unit number, that is, tag along sort number.
Step 1-3-5 is approached according to zero error is carried out to model in step 1-2-4, is provided matrix and be expressed as Heβe=Te,
Wherein, Te=(te1 te2 … tej), tejFor the mark of standard corresponding to Grid Edge ej in training set；G is to select in step 1-3-3
The neuron activation function selected.Pass through solutionObtain output weight.For matrix
Moore-Penrose generalized inverse.So far the quick marking model of Grid Edge has been obtained.
Step 1-4 includes the following steps：
Step 1-4-1, using dough sheet annotation process according to quick dough sheet marking model to the face of threedimensional model in training set
Piece collection carries out component categories mark, and the specific steps are will train the dough sheet feature of lumped model dough sheet collection as input, according to instruction
The quick marking model of the dough sheet got obtains the output valve of model, by output valve by softmax Function Mapping to [0,1],
To obtain with dough sheet feature vector, XfDough sheet f class probability be distributed P (lf,Xf), lfIndicate the mark of patch grids f.
Step 1-4-2, using while annotation process according to quick Grid Edge marking model in training set threedimensional model while
Collection carry out boundary edge classification mark, the specific steps are using model meshes in training set while collection while feature as input, according to instruction
The quick marking model of the Grid Edge got obtains the output valve of model, and the output valve for corresponding to boundary edge mark is passed through
Sigmoid Function Mapping is to [0,1], to obtain with side feature vector, XeBoundary edge e class probability be distributed P (lf≠
lf',Xe), lf'Indicate that the mark of patch grids f', dough sheet f and dough sheet f' are the adjacent dough sheet with a shared side；
Step 1-4-3 constructs graph model for each threedimensional model in training set, and the node of figure is patch grids, figure
In shared between adjacent dough sheet f and dough sheet f', the negative probability logarithm of the classification annotation of dough sheet cuts the number of optimization as figure
According to item Edata(lf,Xf), Edata(lf,Xf)=- ln (P (lf|Xf)), the negative probability logarithm E of Grid Edge classification annotationweight(lf≠
lf',XeWeight of the) × α as side in graph model is cut optimization using multi-tag figure and is calculated for adjusting the smoothness of partitioning boundary
The mark of each patch grids in target three-dimensional after method calculation optimization.The value range of α is [0,10], in this range
Interior, setting step-length is 0.1 progress grid search optimization, is found so that the average segmentation precision of threedimensional model is highest in training set
α is denoted as the optimal smoothing item weight α that figure cuts modelbest。
Step 2 includes the following steps：
Step 2-1, pre-processes target three-dimensional, extracts the dough sheet feature and every grid of each patch grids
While while feature；
Step 2-2, the quick dough sheet marking model and quick Grid Edge marking model obtained using step 1, using dough sheet
Annotation process carries out component categories mark to the dough sheet collection of target three-dimensional, using side annotation process to target three-dimensional
Side collection carries out boundary edge classification mark；
Step 2-3, according to the dough sheet classification annotation probability distribution of target three-dimensional and Grid Edge classification Marking Probability point
Cloth constructs graph model, cuts optimization by multi-tag figure and is split the smooth of boundary, the target three-dimensional face after being optimized
Piece classification annotation results complete the fast automatic mark to target three-dimensional component categories.
Pretreatment is carried out to target three-dimensional described in step 2-1 and includes model normalized and feature extraction,
In, model normalized includes the following steps：
The mass center of target three-dimensional is moved to coordinate origin by step 2-1-1, and the mass center of threedimensional model is by calculating mould
The area weighted average on all vertex obtains in type；
Step 2-1-2, calculate each dough sheet center of target three-dimensional to its mass center Euclidean distance, take it is all it is European away from
From intermediate value complete target three-dimensional by the coordinate of each point in target three-dimensional divided by the specification item as specification item
Normalized；
Feature extraction includes the following steps：
Step 2-1-3, by the dough sheet feature of each patch grids of target three-dimensional, by curvature feature, PCA feature, shape
Shape characteristics of diameters, average geodesic distance feature and Shape context feature cascade to form a feature vector；
Step 2-1-4, by the side feature of every Grid Edge of target three-dimensional, by while dihedral angle feature, while neighborhood vertex
Dihedral angle feature, the curvature difference and derivative feature on side two sides neighborhood vertex, two adjacent surface plate shapes of shared Grid Edge are straight
The shape image difference sign cascade of diameter difference characteristic, two adjacent dough sheets for sharing side forms a feature vector.
Step 2-2 includes the following steps：
Step 2-2-1, using dough sheet annotation process according to quick dough sheet marking model to the dough sheet collection of target three-dimensional
Component categories mark is carried out, the specific steps are using the dough sheet feature of object module dough sheet collection as input, is obtained according to training
The quick marking model of dough sheet obtains the output valve of model, in order to preferably indicate that each dough sheet is labeled as the general of different labels
Rate, by output valve by softmax Function Mapping to [0,1], to obtain with dough sheet feature vector, XfDough sheet f classification it is general
Rate is distributed P (lf,Xf), lfIndicate the mark of patch grids f.
Boundary edge class is carried out according to while collection of the quick Grid Edge marking model to target three-dimensional using in annotation process
It does not mark, the specific steps are using the side feature of object module Grid Edge collection as input, the Grid Edge obtained according to training is quick
Marking model obtains the output valve of model, will correspond to boundary edge mark output valve by sigmoid Function Mapping to [0,
1], to obtain with side feature vector, XeBoundary edge e class probability be distributed P (lf≠lf',Xe), lfIndicate patch grids f
Mark, lf'Indicate the mark of patch grids f', dough sheet f and dough sheet f' are with the adjacent dough sheet in a shared side；
Step 2-2-2 constructs graph model for target three-dimensional, and the node of figure is patch grids, and the side of figure is adjacent
Shared side between dough sheet f and dough sheet f', the negative probability logarithm of the classification annotation of dough sheet cut the data item E of optimization as figuredata(l,
Xf), the negative probability logarithm E of Grid Edge classification annotationweight(lf≠lf',Xe)×αbestAs the weight on side in graph model, it is used for
It is split the smooth optimization on boundary, each net in the target three-dimensional after optimization algorithm calculation optimization is cut using multi-tag figure
The mark l of lattice dough sheet ff。
Beneficial effect：The present invention has the following advantages that：It is trained firstly, the present invention can mark same type of threedimensional model
Collection carries out quickly training study, obtains the quick dough sheet marking model and quick Grid Edge marking model of such threedimensional model；Its
Secondary, the training time of the invention greatly reduces compared to the previous supervised learning time, and the training time can be reduced to from a few houres
Several seconds；Finally, the dough sheet markup information of threedimensional model in training set is not only utilized in the present invention, training is also further utilized
The Grid Edge markup information of threedimensional model is concentrated, to further increase segmentation and mark precision.
Detailed description of the invention
The present invention is done with reference to the accompanying drawings and detailed description and is further illustrated, of the invention is above-mentioned
And/or otherwise advantage will become apparent.
Fig. 1 is processing flow schematic diagram of the invention.
Fig. 2 a is the standard mark of dough sheet.
Fig. 2 b is the standard mark of Grid Edge.
Fig. 3 a is a threedimensional model to be marked.
Fig. 3 b be the present invention by the study of Fig. 2 a training set quick dough sheet marking model to the target three-dimensional of Fig. 3 a into
The result of row dough sheet mark.
Fig. 3 c is target three-dimensional of the quick side marking model of the invention obtained by the training of Fig. 2 b training set to Fig. 3 a
Carry out the result of Grid Edge mark.
Fig. 3 d is that the dough sheet mark that the present invention is learnt by training set and Grid Edge Marking Probability construct graph model, is carried out
Multi-tag figure cut after optimization to the target three-dimensional of Fig. 3 a carry out dough sheet mark as a result, i.e. final segmentation mark knot
Fruit.
Specific embodiment
A kind of quick threedimensional model component categories automatic marking method disclosed by the invention, specifically includes following steps
(flow chart that Fig. 1 is entire method)：
Step 1, threedimensional model mark the quick training of training set：It is that component has been in same class model and model in training set
Provide standard mark, wherein markup information include in threedimensional model each patch grids be subordinated to component parts classification mark with
And every Grid Edge is subordinated to the classification mark of boundary edge.As shown in Figure 2 a, it is marked for the standard of dough sheet.Fig. 2 b is Grid Edge
Standard mark.The different labeled class that each model aircraft patch grids is indicated by different colours is labeled, in training set
Be labeled as standard mark.It is quickly trained by marking training set to threedimensional model, acquisition can be used for unknown three-dimensional mould
Type is split quick dough sheet marking model and boundary marking model with mark.And learns to obtain figure according to training set data and cut
The smooth item weight of model.
Step 2, the mark of target three-dimensional：Quick face obtained by training process using threedimensional model mark training set
Piece marking model and quick Grid Edge marking model are respectively labeled the dough sheet of target three-dimensional and Grid Edge, to obtain
The each dough sheet for obtaining target three-dimensional is subordinated to the classification mark of component parts and every Grid Edge is subordinated to boundary edge
Classification Marking Probability constructs graph model, and cuts optimization by multi-tag figure and obtain smooth segmentation and annotation results.
Lower mask body introduces the main flow of each step：
1, the quick training of threedimensional model mark training set
It is quickly trained by marking training set to threedimensional model, acquisition can be used for being labeled unknown threedimensional model
Quick dough sheet marking model and quick Grid Edge marking model.Located in advance including to threedimensional models all in training set
Reason obtains three steps of quick dough sheet marking model and quick Grid Edge marking model using improved ELM method.
1.1, threedimensional model pre-processes
Threedimensional model preprocessing process includes two steps of model normalized and feature extraction.
Input：Threedimensional model marks training set.
Output：Threedimensional model marks the dough sheet feature of all dough sheets of training set and the side feature of all Grid Edges.
Step 1 pre-processes threedimensional models all in training set, obtains normalized threedimensional model collection.Including first
The mass center of threedimensional model is moved to coordinate origin, the mass center of model is average by the Area-weighted on all vertex on computation model
Value obtains, i.e., the coordinate on each vertex is by the inclusion of all dough sheet areas on the vertex and as weight.Mass center mV(x, y's, z)
It is calculated as：
Wherein, V indicates that the vertex set of threedimensional model, the three-dimensional coordinate of vertex i are (xi,yi,zi), the surface comprising vertex i
Area is ai, VnumFor the vertex number in vertex set.
Then each dough sheet center of threedimensional model is calculated to the Euclidean distance of its mass center, and the intermediate value of all Euclidean distances is taken to make
For specification item, by the coordinate of each point on threedimensional model divided by the specification item, to complete the normalized of threedimensional model.
Step 2 extracts the dough sheet feature of each patch grids of threedimensional model, including curvature feature, PCA feature, shape diameter
Feature, average geodesic distance feature and Shape context feature, cascade form a feature vector.Wherein, curvature feature reference
Document 4：Ohtake Y.,Belyaev A.,Seidel H.-P.Ridge-valley lines on meshes via
implicit surface fitting.ACM Transactions on Graphics,2004,23(3):609-612. described
Method calculates.1 the method for PCA feature reference literature calculates.Shape diameter Function feature reference literature 5：Shapira L.,
Shalom S.,Shamir A.,Cohen-OrD.,ZhangH.Contextual part analogies in 3D
objects.InternationalJournal of Computer Vision,2010,89(2-3):309-326. the method
It calculates.Average geodesic distance feature reference literature 6：Hilaga M.,Shinagawa Y.,Kohmura T.,Kunii
T.L.Topology matching for fully automatic similarity estimationof 3d
shapes.Proceedings of the 28th annual conferenceon Computer graphics and
Described in interactive techniques (New York, NY, USA, 2001), SIGGRAPH ' 01, ACM, pp.203-212.
Method calculates.Shape context feature reference literature 7：Belongie S.,Malik J.,Puzicha J.Shape
matchingand object recognition using shape contexts.IEEE Transactions
OnPattern Analysis and Machine Intelligence,2002,24(4):509-522. the method calculates.
Step 3 extract every Grid Edge of threedimensional model side feature, by while dihedral angle feature, while neighborhood vertex dihedral angle
Feature, while two sides neighborhood vertex curvature difference and derivative feature, it is shared while two adjacent surface plate shape diameter difference features, altogether
The shape image difference feature of two adjacent dough sheets on side is enjoyed, cascade forms a feature vector.The extraction reference of side feature
1 the method for document calculates.
1.2, quick dough sheet marking model
Quick dough sheet marking model acquisition process carries out Fast Classification training to the dough sheet of the threedimensional model in training set, into
And obtain the data item of graph model.
Input：Threedimensional model dough sheet marks training set
Output：The data item of graph model.
Step 1 is according to the dough sheet feature X of threedimensional model dough sheetfIt is instructed with its standard mark l using improved ELM method
Practice.Wherein, l ∈ L is marked, L is predefined all possible dough sheet sub-categories mark set, e.g., " fuselage ", " wing ",
" vertical tail " and " tailplane ".According to the dough sheet collection of threedimensional models all in training set and dough sheet feature Xf, and it is each
The existing standard mark of dough sheet, using document 8：Huang G.-B.,Zhou H.,Ding X.,ZhangR.Extreme
Learning machine for regression and multiclass classification., 2012,42 (2):
513–529.The ELM classification method is trained.During ELM classification based training, the activation primitive of neuron is modified
For ReLu (Rectified Linear Units) function, to be further improved ELM described in original text in stability and precision
Classifying quality, and it is further reduced computing cost.
Classification of the step 2 for model dough sheet to be marked in ELM classifier exports as a result, being obtained by softmax function
With dough sheet feature vector, XfDough sheet, be labeled as l probability distribution P (l | Xf), the corresponding data item of graph model is：
Edata(l,Xf)=- ln (P (l | Xf))。
1.3, quick Grid Edge marking model
Quick Grid Edge marking model acquisition process carries out Fast Classification instruction to the Grid Edge of the threedimensional model in training set
Practice, and then obtains the weight on side in graph model.
Input：Three-dimensional model gridding side marks training set
Output：The side right weight of graph model.
Step 1 according to three-dimensional model gridding while while feature XeIt is quickly trained with its standard mark.For " boundary
For side ", the dough sheet mark for sharing the dough sheet f and dough sheet f' of the Grid Edge is different, i.e. lf≠lf'；Conversely, if sharing the grid
The dough sheet of the dough sheet f and dough sheet f' on side mark identical, i.e. lf≠lf', then it is " non-boundary edge ".According to three-dimensionals all in training set
The Grid Edge collection and side feature X of modeleAnd the existing standard mark in each edge circle side, using ELM method described in document 7
It is trained.During ELM training, activation primitive using ReLu function as neuron, thus in stability, precision
With ELM classifying quality described in further improvement original text in efficiency.
Step 2 is exported in the test of ELM classifier as a result, passing through sigmoid function call for model meshes side to be marked
To the Grid Edge with side feature vector e, the probability distribution P (l of boundary edgef≠lf'|Xe), the corresponding side right weight of graph model is：
Eweight(lf≠lf',Xe)=- ln (P (lf≠lf',Xe))
1.4, learn to obtain figure according to training set data and cut and optimize the optimal weight of smooth item.
Input：The data item and side right weight for the figure that training set is obtained according to dough sheet marking model.
Output：Figure cuts the optimal weight for optimizing smooth item.
Step 1 constructs graph model for each threedimensional model in training set, and the node of figure is patch grids, and the side of figure is
Shared side between adjacent dough sheet f and dough sheet f', data item Edata(lf,Xf), Edata(lf,Xf)=- ln (P (lf|Xf)), smoothly
Item Eweight(lf≠lf',XeWeight of the) × α as side in graph model, for adjusting the smoothness of partitioning boundary, using more marks
Label figure cuts the mark of each patch grids in the target three-dimensional after optimization algorithm calculation optimization.The value range of α be [0,
10], within this range, setting step-length be 0.1 carry out grid search optimization, find so that in training set threedimensional model average mark
The highest α of precision is cut, the optimal smoothing item weight α that figure cuts model is denoted asbest。
2, the mark of target three-dimensional
The resulting quick dough sheet marking model of training process and quick Grid Edge mark using threedimensional model mark training set
Injection molding type carries out the mark of dough sheet and Grid Edge to target three-dimensional, constructs graph model, cuts optimization by multi-tag figure and obtains
The classification of target three-dimensional component parts marks.The target three-dimensional be with training lumped model be subordinated to same kind and
It is constituted with same parts, but not yet divides the threedimensional model with mark, process is as follows：
Input：Target three-dimensional.
Output：Dough sheet mark after target three-dimensional segmentation.
Step 1 pre-processes object module, extracts the dough sheet feature of each patch grids and the side of every Grid Edge
Feature.
Step 2 is using dough sheet annotation process according to quick dough sheet marking model to the dough sheet collection carry out portion of target three-dimensional
Part classification mark carries out boundary edge according to while collection of the quick Grid Edge marking model to target three-dimensional using in annotation process
Classification mark.Wherein, the ELM output valve of dough sheet classification and side classification, is reflected by softmax function and sigmoid function respectively
Be mapped to [0,1], for indicate dough sheet class probability distribution P (l | Xf) and boundary edge probability distribution P (lf≠lf'|Xe)。
Step 3 constructs graph model for target three-dimensional, and figure interior joint is patch grids, and side is adjacent dough sheet f and f'
Between shared side, the classification of dough sheet marks the data item E that negative probability logarithm cuts optimization as figuredata(l,Xf), Grid Edge is subordinated to
The negative probability logarithm E of the mark of boundary edgeweight(lf≠lf',Xe) weight as side in graph model.Using document 9：Boykov
Y.,Veksler O.,Zabih R.Fast Approximate Energy Minimization via Graph
Cuts.IEEE Transactions OnPattern Analysis and Machine Intelligence,2001,23
(11):Multi-tag figure in 1222-1239. cuts optimization algorithm, each patch grids f in the target three-dimensional after calculation optimization
Mark lf。
The target three-dimensional of input is normalized in the preprocessing process, extracts the face of each patch grids
The side feature of piece feature and every Grid Edge, process are as follows：
Input：Target three-dimensional.
Output：The side feature of the dough sheet feature and each edge of each dough sheet in target three-dimensional.
The target of the dough sheet annotation process is that the dough sheet inputted in threedimensional model is initially marked, and is labeled as l ∈ L, L
Gather for predefined all possible mark, e.g., " fuselage ", " wing ", " tailplane " and " vertical tail ", process is such as
Under：
Input：The dough sheet collection of quick dough sheet marking model and target three-dimensional.
Output：The target three-dimensional dough sheet concentrate each dough sheet class probability be distributed P (l | Xf)。
The target of the side annotation process is that the Grid Edge inputted in target three-dimensional is labeled.For " boundary edge "
For, the dough sheet mark for sharing the dough sheet f and dough sheet f' of the Grid Edge is different, i.e. lf≠lf'；Conversely, for " non-boundary edge "
Speech, the dough sheet for sharing the dough sheet f and dough sheet f' of the Grid Edge mark identical, i.e. lf=lf'.Process is as follows：
Input：The Grid Edge collection of quick boundary marking model and target three-dimensional.
Output：Probability distribution P (the l on the target three-dimensional Grid Edge concentration border sidef≠lf'|Xe)
The target of the structure figures model process is to mark to carry out the smooth of boundary and optimization to initial dough sheet, to obtain
The classification of the component parts of target three-dimensional is labeled, process is as follows：
Input：Graph model, the data item E including figure interior jointdata(l,Xf)=- ln (P (l | Xf)), smooth item EsmoothWith
And the weight E on side is connected between figure interior jointweight(lf≠lf',Xe)×αbest=-ln (P (lf≠lf',Xe))×αbest。
Output：Each dough sheet f is subordinated to the mark l of component parts classification in target three-dimensionalf。
Step 1 constructs a figure, and the node of figure is the dough sheet of threedimensional model, and there are Grid Edges as figure between adjacent dough sheet
Side, dough sheet f and dough sheet f' are noted as mark lfWith mark lf'When, side { the weight E of f, f'}weight(lf≠lf',Xe) be：
Eweight(lf≠lf',Xe)×αbest=-ln (P (lf≠lf',Xe))×αbest
Step 2 calculates data item Edata(l,Xf)：
Edata(l,Xf)=- ln (P (l | Xf))
Smooth item E is arranged in step 3smoothIt is 0 for leading diagonal, the L that other elements are 1numRank square matrix, for indicating dough sheet
Classification marks the distance between label measurement.Wherein, LnumClassification number is marked for the standard of dough sheet.
Step 4 cuts optimization algorithm using the multi-tag figure in document 9, by way of calculating the figure minimal cut, calculates three
The best label l of each dough sheet on dimension module grid optimizes graph model, and then acquires after optimization each net in target three-dimensional
The mark l of lattice dough sheet ff。
In the present invention, it is illustrated in figure 2 the threedimensional model mark training set example of input, it is described quick through the invention
Threedimensional model component categories automatic marking method can be split model in Fig. 3 a, obtain initial dough sheet classification chart 3b, side
Class probability Fig. 3 c, construct graph model and carry out figure cut optimization after dough sheet be labeled as shown in Figure 3d.
The present invention provides a kind of threedimensional model component categories automatic marking method, the method for implementing the technical solution
It is many with approach, the above is only a preferred embodiment of the present invention, it is noted that for the ordinary skill of the art
For personnel, various improvements and modifications may be made without departing from the principle of the present invention, these improvements and modifications
It should be regarded as protection scope of the present invention.All undefined components in this embodiment can be implemented in the prior art.
Claims (1)
1. a kind of quick threedimensional model component categories automatic marking method, which is characterized in that include the following steps：
Step 1, training threedimensional model marks training set, and the threedimensional model in training set is to have given with component in class model and model
The quasi- markup information of bid, wherein markup information includes the classification mark that each patch grids is subordinated to component parts in threedimensional model
And every Grid Edge is subordinated to the classification mark of boundary edge, is quickly trained, is obtained by marking training set to threedimensional model
The quick dough sheet marking model and quick Grid Edge marking model with mark must be used to be split target three-dimensional, obtained
Threedimensional model dough sheet classification annotation probability distribution and Grid Edge classification annotation probability distribution, and study obtains figure and cuts the smooth of model
Item weight；
Step 2, label target threedimensional model：The quick dough sheet marking model obtained using step 1 and quick Grid Edge mark mould
Type is split and marks to target three-dimensional respectively, and each dough sheet for obtaining target three-dimensional is subordinated to component parts
Classification mark and every Grid Edge are subordinated to the classification mark of boundary edge, construct graph model, are cut and optimized by multi-tag figure
To smooth segmentation and annotation results；
Step 1 includes the following steps：
Step 1-1 pre-processes threedimensional models all in training set, extract each patch grids dough sheet feature and every
The side feature of Grid Edge；
Step 1-2, all dough sheet composing training patch grids collection of threedimensional model in training set, using improved extreme learning machine
Method is quickly trained the dough sheet feature and its standard mark of each patch grids of training patch grids concentration, is obtained
Quick dough sheet marking model；
Step 1-3, all Grid Edge composing training Grid Edge collection of threedimensional model in training set, using improved extreme learning machine
The side feature and its standard mark for every Grid Edge that method concentrates training Grid Edge are quickly trained, and quick grid is obtained
Side marking model；
Step 1-4 obtains threedimensional model dough sheet classification annotation probability distribution and Grid Edge classification annotation probability distribution, and learns
The optimal smoothing item weight of model is cut to figure；
In step 1-1, described to carry out pretreatment to threedimensional models all in training set include that model normalized and feature mention
It takes, wherein model normalized includes the following steps：
The mass center of threedimensional model in training set is moved to coordinate origin by step 1-1-1, and the mass center of threedimensional model passes through calculating
The area weighted average on all vertex obtains on model；
Step 1-1-2 calculates each dough sheet center of threedimensional model in training set to the Euclidean distance of its mass center, takes all European
The intermediate value of distance completes training set by the coordinate of each point on the threedimensional model in training set divided by the specification item as specification item
In threedimensional model normalized；
Feature extraction includes the following steps：
Step 1-1-3, by the dough sheet feature of each patch grids of threedimensional model in training set, by curvature feature, PCA feature,
Shape characteristics of diameters, average geodesic distance feature and Shape context feature cascade to form a feature vector；
Step 1-1-4, by the side feature of every Grid Edge of threedimensional model in training set, by while dihedral angle feature, while neighborhood top
The dihedral angle feature of point, two adjacent surface plate shapes of the curvature difference and derivative feature on side two sides neighborhood vertex, shared Grid Edge
The shape image difference sign cascade of diameter difference feature, two adjacent dough sheets for sharing side forms a feature vector；
Step 1-2 includes the following steps：
Training patch grids is concentrated the dough sheet feature of each dough sheet as the defeated of Single hidden layer feedforward neural networks by step 1-2-1
Enter cell value to be inputted；
Step 1-2-2, modeling Single hidden layer feedforward neural networks are：
Wherein, j=1,2 ... N, N are the quantity that training set patch grids concentrates dough sheet, xjFor the corresponding dough sheet feature of dough sheet j to
Amount,For the number of hidden layer neuron, wi=(wi1,wi2,…,wiN)TIndicate what hidden layer neuron was connect with input unit
Weight vector,For offset vector, g (y) is activation primitive；ljFor according to face in the resulting training set of model
The reality output of piece j, βkFor the output weight that output unit is connect with hidden layer, k=1,2 ... m, m are the dimension of output layer
Degree, that is, the number of tags classified；
Step 1-2-3, selects ReLu neuron activation function as activation primitive, and ReLu neuron activation function is defined as：G (y)=
max(0,y)；
Step 1-2-4, when approaching the model with zero error,I.e.：
Wherein, tjFor the mark of standard corresponding to dough sheet j in training set, g is the neuron activation function selected in step 1-2-3, hidden
Weight matrix between hiding layer and output layer is expressed as：
Wherein,Indicate the weight vector of output unit and r-th of hidden layer neuron,
Step 1-2-5 is approached according to zero error is carried out to model in step 1-2-4, is provided matrix and be expressed as H β=T, wherein T
=(t1 t2 … tj),Pass through solutionOutput weight is obtained, to obtain quick face
Piece marking model；
Step 1-3 includes the following steps：
Training Grid Edge is concentrated the side feature of each edge as the input unit of Single hidden layer feedforward neural networks by step 1-3-1
Value is inputted；
Step 1-3-2, modeling Single hidden layer feedforward neural networks are：
Wherein, ej=1,2 ... Ne, NeThe quantity of Grid Edge, x are concentrated for training Grid EdgeejFor the corresponding side feature of Grid Edge ej,For the number of hidden layer neuron, wei=(wei1,wei2,…,weiN)TIndicate what hidden layer neuron was connect with input unit
Weight vector,For offset vector, g (y) is activation primitive, lejFor according to the resulting Grid Edge ej's of model
Reality output, βekFor the output weight that output unit is connect with hidden layer, ek=1,2 ... me, meFor the dimension of output layer, i.e.,
The number of tags of classification；
Step 1-3-3, selects ReLu neuron activation function as activation primitive, and ReLu neuron activation function is defined as：G (y)=
max(0,y)；
Step 1-3-4, when approaching the model with zero error,I.e.：
Wherein, tej∈ { 1,2 } is the mark of standard corresponding to Grid Edge ej in training set, corresponds respectively to non-boundary edge and boundary
Side, g are the neuron activation function selected in step 1-3-3, and the weight matrix between hidden layer and output layer is expressed as：
Wherein,Indicate the weight vector of output unit and first of hidden layer neuron, For the quantity of hidden layer neuron, meFor output layer unit number, that is, tag along sort number；
Step 1-3-5 is approached according to zero error is carried out to model in step 1-2-4, is provided matrix and be expressed as Heβe=Te, wherein
Te=(te1 te2 … tej),Pass through solutionOutput weight is obtained, to obtain
Quick Grid Edge marking model；
Step 1-4 includes the following steps：
Step 1-4-1, using dough sheet annotation process according to quick dough sheet marking model to the dough sheet collection of threedimensional model in training set
Component categories mark is carried out, the specific steps are will train the dough sheet feature of lumped model dough sheet collection as input, according to trained
To the quick marking model of dough sheet obtain the output valve of model, by output valve by softmax Function Mapping to [0,1], thus
It obtains with dough sheet feature vector, XfDough sheet f class probability be distributed P (lf,Xf), lfIndicate the mark of patch grids f；
Step 1-4-2, using while annotation process according to quick Grid Edge marking model in training set threedimensional model while collection into
Row bound side classification mark, the specific steps are using model meshes in training set while collection while feature as inputting, according to trained
To the quick marking model of Grid Edge obtain the output valve of model, by correspond to boundary edge mark output valve pass through sigmoid
Function Mapping is to [0,1], to obtain with side feature vector, XeBoundary edge e class probability be distributed P (lf≠lf',Xe),
lf'Indicate that the mark of patch grids f', dough sheet f and dough sheet f' are the adjacent dough sheet with a shared side；
Step 1-4-3 constructs graph model for each threedimensional model in training set, and the node of figure is patch grids, the side of figure
For the shared side between adjacent dough sheet f and dough sheet f', the negative probability logarithm of the classification annotation of dough sheet cuts the data item of optimization as figure
Edata(lf, Xf), Edata(lf, Xf)=- ln (P (lf|Xf)), the negative probability logarithm E of Grid Edge classification annotationweight(lf≠lf',
XeWeight of the) × α as side in graph model cuts optimization algorithm using multi-tag figure for adjusting the smoothness of partitioning boundary
The value range of the mark of each patch grids in target three-dimensional after calculation optimization, weight α is [0,10], in this range
Interior, setting step-length is 0.1 progress grid search optimization, is found so that the average segmentation precision of threedimensional model is highest in training set
Weight α is denoted as the optimal smoothing item weight α that figure cuts modelbest；
Step 2 includes the following steps：
Step 2-1, pre-processes target three-dimensional, extracts the dough sheet feature and every Grid Edge of each patch grids
Side feature；
Step 2-2, the quick dough sheet marking model and quick Grid Edge marking model obtained using step 1, is marked using dough sheet
Process carries out component categories mark to the dough sheet collection of target three-dimensional, using the while collection in annotation process to target three-dimensional
Carry out boundary edge classification mark；
Step 2-3 is distributed, structure according to the dough sheet classification annotation probability distribution of target three-dimensional and Grid Edge classification Marking Probability
Graph model is built, optimization is cut by multi-tag figure and is split the smooth of boundary, the target three-dimensional dough sheet class after being optimized
Other annotation results complete the fast automatic mark to target three-dimensional component categories；
Carrying out pretreatment to target three-dimensional described in step 2-1 includes model normalized and feature extraction, wherein mould
Type normalized includes the following steps：
The mass center of target three-dimensional is moved to coordinate origin by step 2-1-1, and the mass center of threedimensional model passes through on computation model
The area weighted average on all vertex obtains；
Step 2-1-2 calculates each dough sheet center of target three-dimensional to the Euclidean distance of its mass center, takes all Euclidean distances
Intermediate value completes the normalizing of target three-dimensional by the coordinate of each point in target three-dimensional divided by the specification item as specification item
Change processing；
Feature extraction includes the following steps：
Step 2-1-3 is straight by curvature feature, PCA feature, shape by the dough sheet feature of each patch grids of target three-dimensional
Diameter feature, average geodesic distance feature and Shape context feature cascade to form a feature vector；
Step 2-1-4, by the side feature of every Grid Edge of target three-dimensional, by while dihedral angle feature, while neighborhood vertex two
Face angle feature, two adjacent surface plate shape diameter differences of the curvature difference and derivative feature on side two sides neighborhood vertex, shared Grid Edge
The shape image difference sign cascade of different feature, two adjacent dough sheets for sharing side forms a feature vector；
Step 2-2 includes the following steps：
Step 2-2-1 is carried out using dough sheet annotation process according to dough sheet collection of the quick dough sheet marking model to target three-dimensional
Component categories mark, the specific steps are using the dough sheet feature of object module dough sheet collection as inputting, according to trained obtained dough sheet
Quick marking model obtains the output valve of model, by output valve by softmax Function Mapping to [0,1], to be had
Dough sheet feature vector, XfDough sheet f class probability be distributed P (lf,Xf), lfIndicate the mark of patch grids f；
Boundary edge classification mark is carried out according to while collection of the quick Grid Edge marking model to target three-dimensional using in annotation process
Note is quickly marked the specific steps are using the side feature of object module Grid Edge collection as input according to the Grid Edge that training obtains
Model obtains the output valve of model, will correspond to the output valve of boundary edge mark by sigmoid Function Mapping to [0,1], from
And it obtains with side feature vector, XeBoundary edge e class probability be distributed P (lf≠lf',Xe), lf'Indicate patch grids f''s
Mark, dough sheet f and dough sheet f' are the adjacent dough sheet with a shared side；
Step 2-2-2 constructs graph model for target three-dimensional, and the node of figure is patch grids, and the side of figure is adjacent dough sheet f
Shared side between dough sheet f', the negative probability logarithm of the classification annotation of dough sheet cut the data item E of optimization as figuredata(lf,Xf),
Edata(lf,Xf)=- ln (P (lf|Xf)), the negative probability logarithm E of Grid Edge classification annotationweight(lf≠lf',Xe)×αbestAs
The weight on side in graph model, for being split the smooth optimization on boundary, after cutting optimization algorithm calculation optimization using multi-tag figure
Target three-dimensional in each patch grids mark.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610530040.6A CN106157375B (en) | 2016-07-06 | 2016-07-06 | A kind of threedimensional model component categories automatic marking method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610530040.6A CN106157375B (en) | 2016-07-06 | 2016-07-06 | A kind of threedimensional model component categories automatic marking method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106157375A CN106157375A (en) | 2016-11-23 |
| CN106157375B true CN106157375B (en) | 2018-11-30 |
Family
ID=58062820
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610530040.6A Active CN106157375B (en) | 2016-07-06 | 2016-07-06 | A kind of threedimensional model component categories automatic marking method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106157375B (en) |
Families Citing this family (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109255807B (en) * | 2017-07-13 | 2023-02-03 | 腾讯科技（深圳）有限公司 | Image information processing method, server and computer storage medium |
| JP6659641B2 (en) * | 2017-09-13 | 2020-03-04 | ファナック株式会社 | 3D model creation device |
| CN107833209B (en) * | 2017-10-27 | 2020-05-26 | 浙江大华技术股份有限公司 | X-ray image detection method and device, electronic equipment and storage medium |
| CN108304876B (en) * | 2018-01-31 | 2021-07-06 | 国信优易数据股份有限公司 | Classification model training method and device and classification method and device |
| CN108460415B (en) * | 2018-02-28 | 2021-06-15 | 国信优易数据股份有限公司 | Language identification method |
| CN108427970A (en) * | 2018-03-29 | 2018-08-21 | 厦门美图之家科技有限公司 | Picture mask method and device |
| CN108898679B (en) * | 2018-04-10 | 2023-07-21 | 宁波财经学院 | Automatic labeling method for serial numbers of parts |
| CN109199604B (en) * | 2018-08-31 | 2020-12-01 | 浙江大学宁波理工学院 | Multi-feature-based pedicle screw optimal entry point positioning method |
| CN109815830A (en) * | 2018-12-28 | 2019-05-28 | 梦多科技有限公司 | A method of obtaining foot information in the slave photo based on machine learning |
| CN110543625A (en) * | 2019-07-29 | 2019-12-06 | 北京航空航天大学 | rapid marking editing and displaying method for 3D virtual training courseware information |
| CN110533781B (en) * | 2019-08-28 | 2023-07-25 | 南京信息职业技术学院 | Automatic labeling method for multi-class three-dimensional model components |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103021029A (en) * | 2013-01-18 | 2013-04-03 | 南京大学 | Automatic labeling method for three-dimensional model component categories |
| CN103473813A (en) * | 2013-09-18 | 2013-12-25 | 南京大学 | Automatic extracting method for three-dimensional model members |
| CN103971415A (en) * | 2014-05-23 | 2014-08-06 | 南京大学 | Online marking method for three-dimensional model component |
-
2016
- 2016-07-06 CN CN201610530040.6A patent/CN106157375B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103021029A (en) * | 2013-01-18 | 2013-04-03 | 南京大学 | Automatic labeling method for three-dimensional model component categories |
| CN103473813A (en) * | 2013-09-18 | 2013-12-25 | 南京大学 | Automatic extracting method for three-dimensional model members |
| CN103971415A (en) * | 2014-05-23 | 2014-08-06 | 南京大学 | Online marking method for three-dimensional model component |
Non-Patent Citations (3)
| Title |
|---|
| "3D Shape Segmentation and Labeling via Extreme Learning Machine";Zhige Xie et al;《Eurographics Symposium on Geometry Processsing 2014》;20141231;第33卷(第5期);摘要、第1、4.1-4.4.2、7节 * |
| "Low-rank 3D mesh segmentation and labeling with structure guiding";Xiuping Liu et al;《Computers&Graphics》;20141002;99-109 * |
| "Progressive 3D shape segmentation using online learning";Feiqian Zhang et al;《Computer-Aided Design》;20151231;2-12 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106157375A (en) | 2016-11-23 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106157375B (en) | A kind of threedimensional model component categories automatic marking method | |
| Jiao et al. | AF-RCNN: An anchor-free convolutional neural network for multi-categories agricultural pest detection | |
| CN104537676B (en) | Gradual image segmentation method based on online learning | |
| CN102542302B (en) | Automatic complicated target identification method based on hierarchical object semantic graph | |
| CN105825502B (en) | A kind of Weakly supervised method for analyzing image of the dictionary study based on conspicuousness guidance | |
| Shi et al. | Detection and identification of stored-grain insects using deep learning: A more effective neural network | |
| CN108629367A (en) | A method of clothes Attribute Recognition precision is enhanced based on depth network | |
| CN102324038B (en) | Plant species identification method based on digital image | |
| CN103886330A (en) | Classification method based on semi-supervised SVM ensemble learning | |
| CN109002834A (en) | Fine granularity image classification method based on multi-modal characterization | |
| CN109740686A (en) | A kind of deep learning image multiple labeling classification method based on pool area and Fusion Features | |
| CN103971415A (en) | Online marking method for three-dimensional model component | |
| CN103021029A (en) | Automatic labeling method for three-dimensional model component categories | |
| Zhang et al. | Sparse reconstruction for weakly supervised semantic segmentation | |
| WO2024021413A1 (en) | Image segmentation method combining super-pixels and multi-scale hierarchical feature recognition | |
| CN103530633A (en) | Semantic mapping method of local invariant feature of image and semantic mapping system | |
| CN103065158A (en) | Action identification method of independent subspace analysis (ISA) model based on relative gradient | |
| US20200304729A1 (en) | Video processing using a spectral decomposition layer | |
| CN104050460B (en) | The pedestrian detection method of multiple features fusion | |
| CN103473813A (en) | Automatic extracting method for three-dimensional model members | |
| CN105320963B (en) | The semi-supervised feature selection approach of large scale towards high score remote sensing images | |
| Guan et al. | An Object Detection Framework Based on Deep Features and High-Quality Object Locations. | |
| CN106326914A (en) | SVM-based pearl multi-classification method | |
| Zhong et al. | Robust image segmentation against complex color distribution | |
| CN105139452B (en) | A kind of Geologic Curve method for reconstructing based on image segmentation |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |