CN103700094B - The interactive shape collaboration dividing method and device propagated based on label - Google Patents
The interactive shape collaboration dividing method and device propagated based on label Download PDFInfo
- Publication number
- CN103700094B CN103700094B CN201310659606.1A CN201310659606A CN103700094B CN 103700094 B CN103700094 B CN 103700094B CN 201310659606 A CN201310659606 A CN 201310659606A CN 103700094 B CN103700094 B CN 103700094B
- Authority
- CN
- China
- Prior art keywords
- pieces
- sub
- shape
- label
- label information
- 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
Abstract
The present invention is applied to shape collaboration segmentation technology, there is provided a kind of interactive shape collaboration dividing method propagated based on label and device, methods described are included：One group of given shape is too cut into sub-pieces, and builds the relation graph model between sub-pieces；After which part sub-pieces receives the label information that user specifies, based on the relation graph model, the label information marked is propagated in the sub-pieces that other do not mark along the side in figure.Label communications is bound to shape collaboration segmentation field by the present invention, and compared to the collaboration dividing method for being currently based on semi-supervised learning, interactive meanses of the invention are more direct, and faster, accuracy rate is higher for interactive speed；Meanwhile equally efficiently data outside sample can be handled using the inventive method.
Description
Technical field
The invention belongs to shape to cooperate with segmentation technology, more particularly to a kind of interactive shape association propagated based on label
With dividing method and device.
Background technology
Conventional geometric processing method is analyzed 3-D geometric model and handled dependent on local geometric information, but
In recent years, people increasingly have found to be difficult to realize complicated geometric manipulations task just with local geometric information.With research
Work is goed deep into, and people start to excavate and entirety and structural information using geometry, and propose corresponding geometric manipulations
Method, i.e. shape analysis method.Nearly ten years, shape analysis method has obtained extensive concern and hair in geometric manipulations field
Exhibition, turns into present study hotspot.
Recently, in shape analysis process field, scholars think to carry out Cooperative Analysis to multiple shapes, can obtain more
Valuable information, so as to the result being efficiently modified after the processing that performed an analysis to single shape.Shape collaboration segmentation is that this is a kind of
The basis of work, it refers to split one group of shape simultaneously, and establishes the corresponding relation after segmentation between different shape sub-block.Shape
Shape collaboration segmentation can effectively aid in solving many shape process problems, such as modeling, model index, texture mapping.
At present, collaboration dividing method can be roughly divided into three classes：Unsupervised approaches, measure of supervision and semi-supervised method.
In unsupervised approaches, collaboration segmentation can not handle the big situation of dimensional variation, for complexity, or rigid body difference
Property big shape can not robust processing.The block retractility between different shape was defined using style again later, and it is next excellent with this
Method before change, part stretches when can so handle rigid body alignment, but substantially it is also to rely on the rigid body between shape
Alignment.For dependence of the method before overcoming to rigid body requirement, a kind of method of feature based descriptor is occurred, this method is adopted
The similitude between sub-pieces is measured with multiple feature descriptors, and constructs similar matrix.By making feature to the similar matrix
Decompose, most the collaboration segmentation problem of shape sees the clustering problem in spectral space as to this method at last, and obtains for a certain
The collaboration segmentation result of class shape.Because shape description symbols are independently of the position of shape and stretched to wait change, so as to locate
Manage geometry and the bigger data set of change in topology.Multiple descriptors are then made into connection and obtain the similarity measurements between them
Amount.However, the unsupervised approaches of the above all cannot be guaranteed the accuracy of result, the result of segmentation is cooperateed with dependent on given data
Collection.
In supervised learning method, typically split simultaneously using the method for supervised learning and mark shape.It is given
Model that is segmented and having marked, describes these data split and marked, then by these data using descriptor
As training data, a grader is trained.When a given model to be marked, based on the grader, they can be quick
Obtain segmentation and the annotation results of model.If obtained before priori is added in the training data of supervised learning method
The more excellent grader of effect.However, this kind of method needs manually to mark substantial amounts of data as training data, final segmentation and
The result of mark also rely on before training set.
There is scholar to propose a kind of collaboration dividing method based on semi-supervised learning again at present, this method allows user to pass through mark
Note some to constrain in pairs, active assistance participates in collaboration cutting procedure.Demonstrating only needs to specify a small amount of paired constraint, Ta Menneng
Obtain the result of high-accuracy.But this method interactive mode is not very intuitively, it is necessary to which user is designated as to constraint；In addition
The data outside sample can't be handled well, give the data outside a sample, and their methods need to re-execute whole calculation
Method flow.
The content of the invention
In view of the above problems, it is an object of the invention to provide a kind of interactive shape propagated based on label to cooperate with segmentation
Square law device, it is intended to which solving existing shape collaboration splitting scheme needs user to be designated as to constraint, can not handle very well outside sample
The technical problem of data.
On the one hand, the interactive shape collaboration dividing method propagated based on label is comprised the steps：
One group of given shape is too cut into sub-pieces, and builds the relation graph model between sub-pieces；
After which part sub-pieces receives the label information that user specifies, based on the relation graph model, along in figure
Side the label information marked is propagated in the sub-pieces that other do not mark.
On the other hand, the interactive shape collaboration segmenting device propagated based on label is included：
Pretreatment unit, for one group of given shape to be too cut into sub-pieces, and build the relation graph model between sub-pieces；
Label propagation unit, for after which part sub-pieces receives the label information that user specifies, based on the pass
It is graph model, propagates to the label information marked in the sub-pieces that other do not mark along the side in figure.
The beneficial effects of the invention are as follows：Label communications is bound to shape collaboration segmentation field by the present invention, right first
One group of given shape pre-processes through row, including over-segmentation and structure relation graph model, is then interacted with user, Yong Huke
To specify their label information in some sub-pieces, the relation graph model established before is then based on, along the side in figure by
The label information fast propagation of mark cooperates with segmentation result to not marking on node of graph, so as to obtain them.Compared to being currently based on
The collaboration dividing method of semi-supervised learning, interactive meanses of the invention are more direct, and faster, accuracy rate is higher for interactive speed；Meanwhile
Equally efficiently data outside sample can be handled using the inventive method.
Brief description of the drawings
Fig. 1 is the flow chart of the interactive shape collaboration dividing method provided in an embodiment of the present invention propagated based on label；
Fig. 2 is the preferred flow charts of step S101 in Fig. 1；
Fig. 3 is the structure square frame of the interactive shape collaboration segmenting device provided in an embodiment of the present invention propagated based on label
Figure；
Fig. 4 is the structure chart of pretreatment unit provided in an embodiment of the present invention.
Embodiment
In order to make the purpose , technical scheme and advantage of the present invention be clearer, it is right below in conjunction with drawings and Examples
The present invention is further elaborated.It should be appreciated that the specific embodiments described herein are merely illustrative of the present invention, and
It is not used in the restriction present invention.
In order to illustrate technical solutions according to the invention, illustrated below by specific embodiment.
Fig. 1 shows the stream of the interactive shape collaboration dividing method provided in an embodiment of the present invention propagated based on label
Journey, the part related to the embodiment of the present invention is illustrate only for convenience of description.
The interactive shape propagated based on the label collaboration dividing method that the present embodiment provides is included：
Step S101, one group of given shape is too cut into sub-pieces, and builds the relation graph model between sub-pieces.
In graph theory, figure is made up of the side of node and connecting node, and figure is commonly used to describe certain between some things
Particular kind of relationship, things is represented with point, represent that there is this relation between corresponding two things with 2 points of side of connection.In the present embodiment
In, this step pre-processes to one group of given shape first, including over-segmentation and opening relationships graph model.Outside for sample
Geometry（That is the outer data of sample）, can equally be pre-processed using this step.
Step S102, after which part sub-pieces receives the label information that user specifies, based on the relation graph model,
The label information marked is propagated in the sub-pieces that other do not mark along the side in figure.
Need to interact with user after establishing graph model, in this step, user specifies their mark in the sub-pieces of part
Information is signed, is then based on the relation graph model, according to label propagation algorithm, the label information that will have been marked along the side in figure
Propagate in the sub-pieces that other are not marked, segmentation result is cooperateed with so as to obtain them.Compared to the association for being currently based on semi-supervised learning
Same dividing method, the interactive meanses of the present embodiment are more direct, and interactive speed is faster.Meanwhile can be with data outside efficient process sample.
As a kind of preferred embodiment, reference picture 2, above-mentioned steps S101 is specifically included：
Step S201, split by normalizing, by one group of given shape over-segmentation, generate a series of sub-pieces；
Step S202, each sub-pieces is measured with shape description symbols, the similarity measurements of sub-pieces is obtained according to metric
Amount, to construct the relation graph model between sub-pieces, wherein, node in figure represents each sub-pieces, and the side in figure represents two sub-pieces
Similarity measurement.
In this preferred embodiment, shape over-segmentation is generated by a series of sub-pieces using Normalized Cut, then
Relation graph model is built to these sub-pieces, represents the figure in the matrix form here, the node of wherein figure is expressed as each sub-pieces,
The side of figure represents the similarity measurement of two sub-pieces.In order to more accurately measure each sub-pieces, this preferred embodiment selects five
Robust and efficient shape description symbols measure them, so as to preferably distinguish different semantic chunks.The shape description
Accord with as shape diameter function, conformal factor, Shape context, average geodesic distance and the geodesic distance to shaped bases.It is all
These descriptors are all definition and calculated on the dough sheet of grid, i.e., each dough sheet to model, there is five descriptor definitions
Its attribute in some metric space.And each shape is corresponded in each sub-pieces inner sheet to count using histogram
The distribution of shape descriptor.After the measurement for each sub-pieces has been calculated, their phase can be obtained according to these metrics
Measured like property, and construct their relation graph model.
Further, a kind of specific implementation as above-mentioned steps S102, realize that user inputs using iterative algorithm
With label communication process, until obtain user's satisfactory result.Specifically, every time in iteration, figure interior joint along the side in figure to
Label information is propagated around it, while absorbs surroundings nodes and propagates the label information of coming, iterative diffusion is repeated, until all
The label information of node no longer changes, and obtains the collaboration segmentation result of sub-pieces.
Fig. 3 shows the knot of the interactive shape collaboration segmenting device provided in an embodiment of the present invention propagated based on label
Structure, the part related to the embodiment of the present invention is illustrate only for convenience of description.
The interactive shape propagated based on the label collaboration segmenting device that the present embodiment provides is included：
Pretreatment unit 31, for one group of given shape to be too cut into sub-pieces, and build the relation artwork between sub-pieces
Type；
Label propagation unit 32, for after which part sub-pieces receives the label information that user specifies, based on described
Relation graph model, the label information marked is propagated in the sub-pieces that other do not mark along the side in figure.
The functional unit 31,32 that the present embodiment provides, which corresponds to, realizes above-mentioned steps S101, S102, specifically, pre- first
Processing unit 31 is pre-processed to one group of given shape, including over-segmentation and opening relationships graph model, and then label is propagated
Unit 32 is based on the relation graph model, and the label information marked is propagated into other sub-pieces not marked along the side in figure
On.
Preferably, as shown in figure 4, the pretreatment unit 31 includes：
Split module 311, for splitting by normalizing, by one group of given shape over-segmentation, generate a series of sub-pieces；
Module 312 is built, for being measured with shape description symbols to each sub-pieces, the phase of sub-pieces is obtained according to metric
Measured like property, to construct the relation graph model between sub-pieces, wherein, node in figure represents each sub-pieces, and the side in figure represents two
The similarity measurement of individual sub-pieces.
The functional module 311,312 is corresponding to realize above-mentioned steps S201, S202, employs normalization dividing method pair
Shape carries out over-segmentation, obtains a series of sub-pieces, then each sub-pieces is measured with shape description symbols, obtains the phase of sub-pieces
Measured like property, construct the relation graph model between sub-pieces.The shape description symbols are shape diameter function, conformal factor, in shape
Hereafter, average geodesic distance and the geodesic distance to shaped bases.
It is further preferred that above-mentioned label propagation unit 32 includes iterative diffusion module, the iterative diffusion module is used for
After part of nodes receives the label information that user specifies in figure, along the side in figure to node around propagate label information,
While the label information of surroundings nodes propagation is absorbed, until the label information of all nodes no longer changes after iterative diffusion.
To sum up, label communications is bound to shape collaboration segmentation field by the present invention, can obtain interactive mode it is more friendly,
The higher collaboration segmentation result of accuracy rate, at the same it is more efficient to data processing outside sample.
Can be with it will appreciated by the skilled person that realizing that all or part of step in above-described embodiment method is
The hardware of correlation is instructed to complete by program, described program can be stored in a computer read/write memory medium
In, described storage medium, such as ROM/RAM, disk, CD.
The foregoing is merely illustrative of the preferred embodiments of the present invention, is not intended to limit the invention, all essences in the present invention
All any modification, equivalent and improvement made within refreshing and principle etc., should be included in the scope of the protection.
Claims (4)
1. a kind of interactive shape propagated based on label cooperates with dividing method, it is characterised in that methods described includes：
One group of given shape is too cut into sub-pieces, and builds the relation graph model between sub-pieces, one group of given shape
Including the geometry in sample or outside sample；
After which part sub-pieces receives the label information that user specifies, based on the relation graph model, along the side in figure
The label information marked is propagated in the sub-pieces that other do not mark；
It is described that one group of given shape is too cut into sub-pieces, and the graph of a relation model step between sub-pieces is built, specifically include：
Split by normalizing, by one group of given shape over-segmentation, generate a series of sub-pieces；
Each sub-pieces is measured with shape description symbols, and counts pair in each sub-pieces inner sheet using histogram
In the distribution of each shape description symbols, the similarity measurement of sub-pieces should be obtained according to metric, to construct matrix form between sub-pieces
Relation graph model, wherein, node in figure represents each sub-pieces, and the side in figure represents the similarity measurement of two sub-pieces；
It is described after the label information that user specifies is received when which part sub-pieces, based on the relation graph model, along in figure
Side the label information marked is propagated into step in the sub-pieces that other do not mark, specifically include：
After part of nodes receives the label information that user specifies in figure, believe along the side in figure to propagation label around it
Breath, while the label information of surroundings nodes propagation is absorbed, until the label information of all nodes no longer changes after iterative diffusion.
2. method as claimed in claim 1, it is characterised in that the shape description symbols are shape diameter function, conformal factor, shape
Shape context, average geodesic distance and the geodesic distance to shaped bases.
3. a kind of interactive shape propagated based on label cooperates with segmenting device, it is characterised in that described device includes：
Pretreatment unit, for one group of given shape to be too cut into sub-pieces, and the relation graph model between sub-pieces is built, it is described
One group of given shape includes the geometry in sample or outside sample；
Label propagation unit, for after which part sub-pieces receives the label information that user specifies, based on the graph of a relation
Model, the label information marked is propagated in the sub-pieces that other do not mark along the side in figure；
The pretreatment unit includes：
Split module, for splitting by normalizing, by one group of given shape over-segmentation, generate a series of sub-pieces；
Module is built, for measuring each sub-pieces with shape description symbols, and counts each using histogram
The distribution corresponding to each shape description symbols of sub-pieces inner sheet, the similarity measurement of sub-pieces is obtained according to metric, with construction
The relation graph model of matrix form between sub-pieces, wherein, node in figure represents each sub-pieces, and the side in figure represents two sub-pieces
Similarity measurement；
The label propagation unit includes：
Iterative diffusion module, for after part of nodes receives the label information that user specifies in figure, along the side in figure to
Label information is propagated around node, while absorbs the label information of surroundings nodes propagation, up to all nodes after iterative diffusion
Label information no longer changes.
4. device as claimed in claim 3, it is characterised in that the shape description symbols are shape diameter function, conformal factor, shape
Shape context, average geodesic distance and the geodesic distance to shaped bases.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310659606.1A CN103700094B (en) | 2013-12-09 | 2013-12-09 | The interactive shape collaboration dividing method and device propagated based on label |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310659606.1A CN103700094B (en) | 2013-12-09 | 2013-12-09 | The interactive shape collaboration dividing method and device propagated based on label |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN103700094A CN103700094A (en) | 2014-04-02 |
| CN103700094B true CN103700094B (en) | 2017-12-01 |
Family
ID=50361613
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201310659606.1A Active CN103700094B (en) | 2013-12-09 | 2013-12-09 | The interactive shape collaboration dividing method and device propagated based on label |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103700094B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106934489B (en) * | 2017-02-22 | 2020-10-23 | 南京邮电大学 | Time sequence link prediction method for complex network |
| CN109255791A (en) * | 2018-07-19 | 2019-01-22 | 杭州电子科技大学 | A kind of shape collaboration dividing method based on figure convolutional neural networks |
| CN110443814B (en) * | 2019-07-30 | 2022-12-27 | 北京百度网讯科技有限公司 | Loss assessment method, device, equipment and storage medium for vehicle |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101295360A (en) * | 2008-05-07 | 2008-10-29 | 清华大学 | Semi-supervision image classification method based on weighted graph |
Family Cites Families (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2001195599A (en) * | 2000-01-14 | 2001-07-19 | Nippon Telegr & Teleph Corp <Ntt> | Method and device for processing image, and image processing program storage medium |
| US8218870B2 (en) * | 2010-02-18 | 2012-07-10 | Mitsubishi Electric Research Laboratories, Inc. | Method for segmenting images with intensity-based label propagation and probabilistic of level sets |
| CN101853400B (en) * | 2010-05-20 | 2012-09-26 | 武汉大学 | Multiclass image classification method based on active learning and semi-supervised learning |
| CN102495865B (en) * | 2011-11-28 | 2013-08-07 | 南京大学 | Image annotation method combined with image internal space relation and visual symbiosis relation |
| CN103093248B (en) * | 2013-01-28 | 2016-03-23 | 中国科学院自动化研究所 | A kind of semi-supervision image classification method based on various visual angles study |
-
2013
- 2013-12-09 CN CN201310659606.1A patent/CN103700094B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101295360A (en) * | 2008-05-07 | 2008-10-29 | 清华大学 | Semi-supervision image classification method based on weighted graph |
Also Published As
| Publication number | Publication date |
|---|---|
| CN103700094A (en) | 2014-04-02 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Branson et al. | Strong supervision from weak annotation: Interactive training of deformable part models | |
| WO2022001623A1 (en) | Image processing method and apparatus based on artificial intelligence, and device and storage medium | |
| CN100407798C (en) | Three-dimensional geometric mode building system and method | |
| Liu et al. | The research of virtual face based on Deep Convolutional Generative Adversarial Networks using TensorFlow | |
| CN110674741B (en) | Gesture recognition method in machine vision based on double-channel feature fusion | |
| CN105608690B (en) | A kind of image partition method being combined based on graph theory and semi-supervised learning | |
| CN107066935A (en) | Hand gestures method of estimation and device based on deep learning | |
| Lu et al. | PointNGCNN: Deep convolutional networks on 3D point clouds with neighborhood graph filters | |
| Wu et al. | Interactive shape co-segmentation via label propagation | |
| US20190087964A1 (en) | Method and apparatus for parsing and processing three-dimensional cad model | |
| Yang et al. | 3d segmentation of glial cells using fully convolutional networks and k-terminal cut | |
| JP7158563B2 (en) | Deep model training method and its device, electronic device and storage medium | |
| CN104658022B (en) | Three-dimensional animation manufacturing method and device | |
| CN105354593B (en) | A kind of threedimensional model sorting technique based on NMF | |
| CN104200240A (en) | Sketch retrieval method based on content adaptive Hash encoding | |
| CN103700094B (en) | The interactive shape collaboration dividing method and device propagated based on label | |
| CN104103093B (en) | Three-dimensional grid semantic marking method based on deep convolution neural network | |
| CN102938157B (en) | A kind of stylized Freehandhand-drawing based on disk B-spline curve generates method and system | |
| CN112966390B (en) | Method and apparatus for garment processing based on dual three-dimensional distance fields | |
| Chen et al. | A thinning-based liver vessel skeletonization method | |
| CN111814812A (en) | Modeling method, modeling device, storage medium, electronic device and scene recognition method | |
| CN103077404B (en) | Local regularity formula based on markov random file visualizes clustering method | |
| Hu et al. | CVT-based 3D image segmentation and quality improvement of tetrahedral/hexahedral meshes using anisotropic Giaquinta-Hildebrandt operator | |
| Jeong et al. | Artificial intelligence for the fourth industrial revolution | |
| CN102855624A (en) | Image segmentation method based on generalized data field and Normalized cut (Ncut) algorithm |
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 |