CN103700094A - Interactive collaborative shape segmentation method and device based on label propagation - Google Patents
Interactive collaborative shape segmentation method and device based on label propagation Download PDFInfo
- Publication number
- CN103700094A CN103700094A CN201310659606.1A CN201310659606A CN103700094A CN 103700094 A CN103700094 A CN 103700094A CN 201310659606 A CN201310659606 A CN 201310659606A CN 103700094 A CN103700094 A CN 103700094A
- Authority
- CN
- China
- Prior art keywords
- sub
- sheet
- shape
- label information
- label
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Abstract
The invention is applicable to the technical field of collaborative shape segmentation and provides an interactive collaborative shape segmentation method and a device based on label propagation. The method comprises the steps of over-segmenting a group of given shapes into sub-segments, constructing a relational graph model among the sub-segments, and propagating marked label information to the other unmarked sub-segments along the sides in a graph based on the relational graph model after a part of sub-segments receive the label information specified by a user. According to the method and the device, a label propagation technology is combined to the collaborative shape segmentation field; compared with the current collaborative segmentation method based on half-supervision learning, an interactive means of the method and the device is more direct; the interactive speed and the accuracy rate are higher; and at the same time, with the adoption of the method, out-of-sample data can also be processed efficiently.
Description
Technical field
The invention belongs to the collaborative segmentation technology of shape, relate in particular to the collaborative dividing method of a kind of interactive shape of propagating based on label and device.
Background technology
Conventional geometric disposal route depends on local geometric information 3-D geometric model is analyzed and processed, yet in recent years, people more and more find only to utilize local geometric information to be difficult to realize complicated geometric manipulations task.Along with going deep into of research work, people start to excavate and utilize integral body and the structural information of geometric configuration, and have proposed corresponding geometric manipulations method, i.e. shape analysis method.Nearly ten years, shape analysis method has obtained paying close attention to widely and developing in geometric manipulations field, becomes present study hotspot.
Recently, in shape analysis process field, scholars think a plurality of shapes are carried out to Cooperative Analysis, can obtain more valuable information, thereby effectively improve the result that single shape is performed an analysis after processing.Shape collaborative cutting apart is the basis of this class work, and it refers to cuts apart one group of shape simultaneously, and sets up and cut apart the corresponding relation between rear difformity sub-block.Shape can effectively assist a lot of shapes of solution to process problems collaborative cutting apart, as modeling, model index, texture etc.
At present, collaborative dividing method can roughly be divided three classes: without measure of supervision, measure of supervision and semi-supervised method.
In without measure of supervision, can not process the situation that dimensional variation is large collaborative cutting apart, for complexity, or the large shape of rigid body otherness can not be processed by robust.Adopted again afterwards style to define the piece retractility between difformity, and the method before optimizing with this, in the time of processing like this rigid body alignment, parts are flexible, but it still depends on the rigid body alignment between shape in essence.For the dependence of method to rigid body requirement before overcoming, occurred a kind of method based on feature descriptor, the method adopts a plurality of feature descriptors to measure the similarity between sub-sheet, and constructs similar matrix.By this similar matrix is made to feature decomposition, the method the most collaborative segmentation problem of shape is seen the clustering problem in spectral space as, and obtains the collaborative segmentation result for a certain class shape.Because shape description symbols is the position and flexible grade that are independent of shape, change, thereby can process how much and the larger data set of change in topology.Subsequently a plurality of descriptors are connected to the similarity measurement obtaining between them.Yet above all can not guarantee the accuracy of result without measure of supervision, the collaborative result of cutting apart depends on given data set.
In supervised learning method, generally adopt the method for supervised learning to cut apart simultaneously and mark shape.The more given model of having cut apart and having marked, is used descriptor to describe the data that these have been cut apart and have marked, and then using these data as training data, trains a sorter.When a given model to be marked, based on this sorter, they can obtain cutting apart and annotation results of model fast.If in the training data of supervised learning method, obtain the more excellent sorter of effect before priori is added.Yet this class methods need a large amount of data of artificial mark as training data, finally cut apart and the result that mark also depends on previous training set.
At present have again scholar to propose a kind of collaborative dividing method based on semi-supervised learning, the method allows user to retrain in pairs by marking some, and active assistance participates in collaborative cutting procedure.Proved and only needed to specify a small amount of paired constraint, they can obtain the result of high-accuracy.But this method interactive mode is not very directly perceived, need user to be designated as constraint; Can't well process the data outside sample in addition, the data outside a given sample, they need to re-execute whole algorithm flow at method.
Summary of the invention
In view of the above problems, the object of the present invention is to provide the collaborative dividing method device of a kind of interactive shape of propagating based on label, be intended to solve the collaborative splitting scheme of existing shape need user be designated as to constraint, can not the outer data of fine processing sample technical matters.
On the one hand, the collaborative dividing method of the described interactive shape of propagating based on label comprises the steps:
One group of given shape is too slit into sub-sheet, and builds the graph of a relation model between sub-sheet;
When parton sheet wherein receives after the label information of user's appointment, based on described graph of a relation model, the limit in figure is transmitted to the label information having marked on other sub-sheets that do not mark.
On the other hand, the collaborative segmenting device of the described interactive shape of propagating based on label comprises:
Pretreatment unit, for one group of given shape is too slit into sub-sheet, and builds the graph of a relation model between sub-sheet;
Label propagation unit, for receiving after the label information of user's appointment when parton sheet wherein, based on described graph of a relation model, the limit in figure is transmitted to the label information having marked on other sub-sheets that do not mark.
The invention has the beneficial effects as follows: the present invention is bonded to the collaborative field of cutting apart of shape by label communications, first to one group of given shape through row pre-service, comprise over-segmentation and build graph of a relation model, then carry out alternately with user, user can specify their label information on one a little, then the graph of a relation model based on setting up before, does not extremely mark the label information fast propagation having marked on node of graph along the limit in figure, thereby obtains their collaborative segmentation results.Compare the collaborative dividing method based on semi-supervised learning at present, interactive means of the present invention is more direct, and interactive speed is faster, and accuracy rate is higher; Meanwhile, adopt the inventive method can to the outer data of sample, process efficiently equally.
Accompanying drawing explanation
Fig. 1 is the process flow diagram of the collaborative dividing method of the interactive shape of propagating based on label that provides of the embodiment of the present invention;
Fig. 2 is the preferred flow charts of step S101 in Fig. 1;
Fig. 3 is the block diagram of the collaborative segmenting device of the interactive shape of propagating based on label that provides of the embodiment of the present invention;
Fig. 4 is the structural drawing of the pretreatment unit that provides of the embodiment of the present invention.
Embodiment
In order to make object of the present invention, technical scheme and advantage clearer, below in conjunction with drawings and Examples, the present invention is further elaborated.Should be appreciated that specific embodiment described herein, only in order to explain the present invention, is not intended to limit the present invention.
For technical solutions according to the invention are described, below by specific embodiment, describe.
The interactive shape of propagating based on label that Fig. 1 shows the embodiment of the present invention to be provided is worked in coordination with the flow process of dividing method, only shows for convenience of explanation the part relevant to the embodiment of the present invention.
The collaborative dividing method of the interactive shape of propagating based on label that the present embodiment provides comprises:
Step S101, one group of given shape is too slit into sub-sheet, and builds the graph of a relation model between sub-sheet.
In graph theory, figure consists of the limit of node and connected node, and figure is commonly used to describe certain particular kind of relationship between some things, with point, represents things, with the limit that connects at 2, represents to have this relation between corresponding two things.In the present embodiment, first this step carries out pre-service to one group of given shape, comprises over-segmentation and opening relationships graph model.For the geometric configuration outside sample (being the outer data of sample), can adopt equally this step to carry out pre-service.
Step S102, when parton sheet wherein receives after the label information of user's appointment, based on described graph of a relation model, the limit in figure is transmitted to the label information having marked on other sub-sheets that do not mark.
Set up after graph model, in this step, need to carry out alternately with user, user specifies their label information on parton sheet, then based on described graph of a relation model, according to label propagation algorithm, limit in figure is transmitted to the label information having marked on other sub-sheets that do not mark, thereby obtains their collaborative segmentation results.Compare the collaborative dividing method based on semi-supervised learning at present, the interactive means of the present embodiment is more direct, and interactive speed is faster.Meanwhile, can efficiently process the outer data of sample.
As a kind of preferred implementation, with reference to Fig. 2, above-mentioned steps S101 specifically comprises:
Step S201, by normalization, cut apart, one group of shape over-segmentation by given, generates a series of sub-sheets;
Step S202, with shape description symbols, every height sheet is measured, obtained the similarity measurement of sub-sheet according to metric, with the graph of a relation model between constructor sheet, wherein, the node in figure represents each sub-sheet, and the limit in figure represents the similarity measurement of two sub-sheets.
In this preferred implementation, adopt Normalized Cut that shape over-segmentation is generated to a series of sub-sheet, then to this little, build graph of a relation model, with matrix form, represent this figure here, wherein the node table of figure is shown every height sheet, and the limit of figure represents the similarity measurement of two sub-sheets.In order to measure more accurately every height sheet, this optimal way selects five robusts and efficient shape description symbols to measure them, thereby can better distinguish different semantic chunks.Described shape description symbols is shape diameter function, the conformal factor, Shape context, average geodesic distance and the geodesic distance that arrives shape bottom.All these descriptors are all definition and calculate on the dough sheet of grid, i.e. each dough sheet to model has its attribute in certain metric space of five descriptor definition.And adopt histogram to add up the distribution corresponding to each shape description symbols that drops on dough sheet in every height sheet.After the tolerance of having calculated for every height sheet, can obtain according to these metrics their similarity measurement, and construct their graph of a relation model.
Further, as a kind of specific implementation of above-mentioned steps S102, adopt iterative algorithm to realize user's input and label communication process, until obtain user's satisfactory result.Concrete, in each iteration, in figure, node is propagated label information around along the Bian Xiangqi in figure, absorb node around simultaneously and propagate the label information of coming, repeat iteration and propagate, until the label information of all nodes no longer changes, obtain the collaborative segmentation result of sub-sheet.
The interactive shape of propagating based on label that Fig. 3 shows the embodiment of the present invention to be provided is worked in coordination with the structure of segmenting device, only shows for convenience of explanation the part relevant to the embodiment of the present invention.
The collaborative segmenting device of the interactive shape of propagating based on label that the present embodiment provides comprises:
Pretreatment unit 31, for one group of given shape is too slit into sub-sheet, and builds the graph of a relation model between sub-sheet;
Label propagation unit 32, for receiving after the label information of user's appointment when parton sheet wherein, based on described graph of a relation model, the limit in figure is transmitted to the label information having marked on other sub-sheets that do not mark.
Functional unit 31,32 correspondences that the present embodiment provides have realized above-mentioned steps S101, S102, concrete, first 31 pairs of one group of given shapes of pretreatment unit are carried out pre-service, comprise over-segmentation and opening relationships graph model, then label propagation unit 32 is based on described graph of a relation model, and the limit in figure is transmitted to the label information having marked on other sub-sheets that do not mark.
Preferably, as shown in Figure 4, described pretreatment unit 31 comprises:
Cut apart module 311, for cutting apart by normalization, one group of shape over-segmentation by given, generates a series of sub-sheets;
Build module 312, for shape description symbols, every height sheet being measured, according to metric, obtain the similarity measurement of sub-sheet, with the graph of a relation model between constructor sheet, wherein, the node in figure represents each sub-sheet, and the limit in figure represents the similarity measurement of two sub-sheets.
Described functional module 311,312 correspondences have realized above-mentioned steps S201, S202, adopted normalization dividing method to carry out over-segmentation to shape, obtain a series of sub-sheets, then with shape description symbols, every height sheet is measured, obtain the similarity measurement of sub-sheet, the graph of a relation model between constructor sheet.Described shape description symbols is shape diameter function, the conformal factor, Shape context, average geodesic distance and the geodesic distance that arrives shape bottom.
Further preferred, above-mentioned label propagation unit 32 comprises iteration propagation module, described iteration propagation module is for receiving after the label information of user's appointment when figure part of nodes, limit in figure is propagated label information around to node, absorb the label information that node is propagated around simultaneously, after iteration is propagated until the label information of all nodes no longer change.
To sum up, the present invention is bonded to label communications that shape is collaborative cuts apart field, can obtain the collaborative segmentation result that interactive mode is more friendly, accuracy rate is higher, simultaneously more efficient to the outer data processing of sample.
One of ordinary skill in the art will appreciate that, the all or part of step realizing in above-described embodiment method is to come the hardware that instruction is relevant to complete by program, described program can be in being stored in a computer read/write memory medium, described storage medium, as ROM/RAM, disk, CD etc.
The foregoing is only preferred embodiment of the present invention, not in order to limit the present invention, all any modifications of doing within the spirit and principles in the present invention, be equal to and replace and improvement etc., within all should being included in protection scope of the present invention.
Claims (8)
1. the collaborative dividing method of interactive shape of propagating based on label, is characterized in that, described method comprises:
One group of given shape is too slit into sub-sheet, and builds the graph of a relation model between sub-sheet;
When parton sheet wherein receives after the label information of user's appointment, based on described graph of a relation model, the limit in figure is transmitted to the label information having marked on other sub-sheets that do not mark.
2. method as claimed in claim 1, is characterized in that, described one group of given shape is too slit into sub-sheet, and builds the graph of a relation model step between sub-sheet, specifically comprises:
By normalization, cut apart, one group of shape over-segmentation by given, generates a series of sub-sheets;
With shape description symbols, every height sheet is measured, obtained the similarity measurement of sub-sheet according to metric, with the graph of a relation model between constructor sheet, wherein, the node in figure represents each sub-sheet, and the limit in figure represents the similarity measurement of two sub-sheets.
3. method as claimed in claim 2, is characterized in that, described shape description symbols is shape diameter function, the conformal factor, Shape context, average geodesic distance and to the geodesic distance of shape bottom.
4. method as claimed in claim 3, it is characterized in that, describedly receive after the label information of user's appointment when parton sheet wherein, based on described graph of a relation model, limit in figure is transmitted to step on other sub-sheets that do not mark by the label information having marked, and specifically comprises:
In figure, part of nodes receives after the label information of user's appointment, along the Bian Xiangqi in figure, propagates label information around, absorbs the label information that node is propagated around simultaneously, after iteration is propagated until the label information of all nodes no longer change.
5. the collaborative segmenting device of interactive shape of propagating based on label, is characterized in that, described device comprises:
Pretreatment unit, for one group of given shape is too slit into sub-sheet, and builds the graph of a relation model between sub-sheet;
Label propagation unit, for receiving after the label information of user's appointment when parton sheet wherein, based on described graph of a relation model, the limit in figure is transmitted to the label information having marked on other sub-sheets that do not mark.
6. install as claimed in claim 5, it is characterized in that, described pretreatment unit comprises:
Cut apart module, for cutting apart by normalization, one group of shape over-segmentation by given, generates a series of sub-sheets;
Build module, for shape description symbols, every height sheet being measured, according to metric, obtain the similarity measurement of sub-sheet, with the graph of a relation model between constructor sheet, wherein, the node in figure represents each sub-sheet, and the limit in figure represents the similarity measurement of two sub-sheets.
7. install as claimed in claim 6, it is characterized in that, described shape description symbols is shape diameter function, the conformal factor, Shape context, average geodesic distance and the geodesic distance that arrives shape bottom.
8. install as claimed in claim 7, it is characterized in that, described label propagation unit comprises:
Iteration propagation module, for receiving after the label information of user's appointment when figure part of nodes, along the limit in figure to node, propagate label information around, absorb the label information that node is propagated simultaneously around, after iteration is propagated until the label information of all nodes no longer change.
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 true CN103700094A (en) | 2014-04-02 |
CN103700094B 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) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106934489A (en) * | 2017-02-22 | 2017-07-07 | 南京邮电大学 | A kind of sequential link Forecasting Methodology towards 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 |
CN110443814A (en) * | 2019-07-30 | 2019-11-12 | 北京百度网讯科技有限公司 | Damage identification method, device, equipment and the storage medium of vehicle |
Citations (6)
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 |
CN101295360A (en) * | 2008-05-07 | 2008-10-29 | 清华大学 | Semi-supervision image classification method based on weighted graph |
CN101853400A (en) * | 2010-05-20 | 2010-10-06 | 武汉大学 | Multiclass image classification method based on active learning and semi-supervised learning |
JP2011170846A (en) * | 2010-02-18 | 2011-09-01 | Mitsubishi Electric Research Laboratories Inc | Method for segmenting image by intensity-based label propagation and at probability of level set |
CN102495865A (en) * | 2011-11-28 | 2012-06-13 | 南京大学 | Image annotation method combined with image internal space relation and visual symbiosis relation |
CN103093248A (en) * | 2013-01-28 | 2013-05-08 | 中国科学院自动化研究所 | Semi-supervised image classification method based on multi-view study |
-
2013
- 2013-12-09 CN CN201310659606.1A patent/CN103700094B/en active Active
Patent Citations (6)
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 |
CN101295360A (en) * | 2008-05-07 | 2008-10-29 | 清华大学 | Semi-supervision image classification method based on weighted graph |
JP2011170846A (en) * | 2010-02-18 | 2011-09-01 | Mitsubishi Electric Research Laboratories Inc | Method for segmenting image by intensity-based label propagation and at probability of level set |
CN101853400A (en) * | 2010-05-20 | 2010-10-06 | 武汉大学 | Multiclass image classification method based on active learning and semi-supervised learning |
CN102495865A (en) * | 2011-11-28 | 2012-06-13 | 南京大学 | Image annotation method combined with image internal space relation and visual symbiosis relation |
CN103093248A (en) * | 2013-01-28 | 2013-05-08 | 中国科学院自动化研究所 | Semi-supervised image classification method based on multi-view study |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106934489A (en) * | 2017-02-22 | 2017-07-07 | 南京邮电大学 | A kind of sequential link Forecasting Methodology towards 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 |
CN110443814A (en) * | 2019-07-30 | 2019-11-12 | 北京百度网讯科技有限公司 | Damage identification method, device, equipment and the storage medium of vehicle |
Also Published As
Publication number | Publication date |
---|---|
CN103700094B (en) | 2017-12-01 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Lu et al. | PHA: A fast potential-based hierarchical agglomerative clustering method | |
CN105512289A (en) | Image retrieval method based on deep learning and Hash | |
JP7110493B2 (en) | Deep model training method and its device, electronic device and storage medium | |
JP7158563B2 (en) | Deep model training method and its device, electronic device and storage medium | |
CN108304380B (en) | Method for disambiguating names of scholars by fusing academic influence | |
CN112580362B (en) | Visual behavior recognition method, system and computer readable medium based on text semantic supervision | |
CN103425757A (en) | Cross-medial personage news searching method and system capable of fusing multi-mode information | |
CN109582799A (en) | The determination method, apparatus and electronic equipment of knowledge sample data set | |
CN103268635A (en) | Segmentation and semantic annotation method of geometry grid scene model | |
CN104200240A (en) | Sketch retrieval method based on content adaptive Hash encoding | |
CN104933029A (en) | Text image joint semantics analysis method based on probability theme model | |
CN103440274A (en) | Video event sketch construction and matching method based on detail description | |
CN110188359B (en) | Text entity extraction method | |
CN103793507A (en) | Method for obtaining bimodal similarity measure with deep structure | |
CN103700094A (en) | Interactive collaborative shape segmentation method and device based on label propagation | |
CN111144109B (en) | Text similarity determination method and device | |
CN102024067A (en) | Method for technology transplant of analog circuit | |
CN108304519A (en) | A kind of knowledge forest construction method based on chart database | |
CN105631465A (en) | Density peak-based high-efficiency hierarchical clustering method | |
Lu et al. | Organizational Graph Generation for Structured Architectural Floor Plan Dataset | |
CN104598937A (en) | Recognizing method and device for text information | |
CN110347841A (en) | A kind of method, apparatus, storage medium and the electronic equipment of document content classification | |
CN102270343B (en) | Image segmentation method based on Ising graph model | |
CN113010677A (en) | Paper collective classification method, system and medium based on Markov chain model | |
CN106802940A (en) | A kind of method and device for calculating text subject model |
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 |