CN106960032B - Three-dimensional shape expression method and device - Google Patents
Three-dimensional shape expression method and device Download PDFInfo
- Publication number
- CN106960032B CN106960032B CN201710171841.2A CN201710171841A CN106960032B CN 106960032 B CN106960032 B CN 106960032B CN 201710171841 A CN201710171841 A CN 201710171841A CN 106960032 B CN106960032 B CN 106960032B
- Authority
- CN
- China
- Prior art keywords
- dimensional shape
- substructures
- dimensional
- substructure
- skeleton
- 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
- 230000014509 gene expression Effects 0.000 title claims abstract description 48
- 238000000034 method Methods 0.000 title claims abstract description 36
- 230000011218 segmentation Effects 0.000 claims abstract description 31
- 238000005070 sampling Methods 0.000 claims description 27
- 238000010586 diagram Methods 0.000 claims description 15
- 239000013598 vector Substances 0.000 claims description 15
- 230000008569 process Effects 0.000 claims description 10
- 238000000605 extraction Methods 0.000 claims description 8
- 230000003993 interaction Effects 0.000 claims description 4
- 230000003595 spectral effect Effects 0.000 claims description 2
- 230000006870 function Effects 0.000 description 4
- 238000004364 calculation method Methods 0.000 description 3
- 230000006872 improvement Effects 0.000 description 3
- 238000000513 principal component analysis Methods 0.000 description 3
- SLXKOJJOQWFEFD-UHFFFAOYSA-N 6-aminohexanoic acid Chemical compound NCCCCCC(O)=O SLXKOJJOQWFEFD-UHFFFAOYSA-N 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004590 computer program Methods 0.000 description 1
- 230000001186 cumulative effect Effects 0.000 description 1
- 238000005315 distribution function Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 230000000717 retained effect Effects 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/58—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
- G06F16/583—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content
- G06F16/5854—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content using shape and object relationship
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Library & Information Science (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Image Analysis (AREA)
- Apparatus For Radiation Diagnosis (AREA)
Abstract
The embodiment of the invention provides a three-dimensional shape expression method and a three-dimensional shape expression device. The method comprises the following steps: extracting a three-dimensional shape mixed skeleton; obtaining the segmentation of the three-dimensional shape by segmenting the mixed type skeleton; obtaining a substructure of the three-dimensional shape according to the segmented three-dimensional shape; and establishing the expression of the three-dimensional shape by utilizing a bag-of-words model according to the substructure of the three-dimensional shape. The embodiment of the invention can realize concise and efficient expression of the three-dimensional shape.
Description
Technical Field
The invention relates to a graphics technology, in particular to a three-dimensional shape expression method and a three-dimensional shape expression device.
Background
The gradual improvement of the geometric information acquisition equipment of the three-dimensional shapes and the maturity of the three-dimensional modeling mode greatly improve the number of the three-dimensional shapes, and further provide higher requirements for the retrieval and the comparison of the three-dimensional shapes. Based on such observations, how to efficiently express a three-dimensional shape is a key to solve the problem.
In recent years, there has been an increasing number of studies relating to the expression of three-dimensional shapes and shape search using different three-dimensional shape expressions. In the prior art, the three-dimensional shape is generally expressed by global or local features, and different feature descriptors are used to describe the three-dimensional shape, for example, the volume, area, fourier transform coefficient and other statistical data of the three-dimensional shape are used as global features to describe the three-dimensional shape, or different three-dimensional shapes are expressed based on the distribution of distance, angle, area and volume between random surface points in the three-dimensional shape, and so on.
In summary, the expression modes of three-dimensional shapes in the existing schemes can be mainly classified into three main categories: (1) a feature-based expression; (2) graph-based expressions; (3) perspective-based representation. The expression based on the characteristics only considers the geometric properties of the surface of the shape and does not consider the integral structure of the three-dimensional shape, and people usually feel the shape structurally but not in detail. The graph-based approach merely represents a three-dimensional shape by connecting graphs, and additional definition and calculation are often required for further comparison or search applications. However, for the expression based on the view angle, the core idea is to capture each view angle of the three-dimensional shape by using a plurality of two-dimensional images, but a large number of two-dimensional images are often required in order to capture the information of the three-dimensional shape as comprehensively as possible. If the number of the two-dimensional pictures is too small, many details on the shape are lost, and the shape expression result is affected. Moreover, while obtaining these two-dimensional images, it is necessary to calculate specific descriptors for these two-dimensional images, and it is not easy to directly apply these images to various applications of three-dimensional shapes.
Therefore, the existing expression modes of three-dimensional shapes have certain problems, and the space for improvement is provided.
Disclosure of Invention
The embodiment of the invention provides a three-dimensional shape expression method and a device, which can express a three-dimensional shape in a simple and efficient manner.
The embodiment of the invention provides a three-dimensional shape expression method, which comprises the following steps: extracting a three-dimensional shape mixed skeleton; obtaining the segmentation of the three-dimensional shape by segmenting the mixed type skeleton; obtaining a substructure of the three-dimensional shape according to the segmented three-dimensional shape; and establishing the expression of the three-dimensional shape by utilizing a bag-of-words model according to the substructure of the three-dimensional shape.
Wherein, the extracting the three-dimensional shape of the hybrid skeleton comprises: sampling the surface of the three-dimensional shape to obtain a sampling point; and re-expressing the sampling points to obtain a mixed skeleton containing a one-dimensional curve and a two-dimensional slice.
Wherein the obtaining of the segmentation of the three-dimensional shape by the segmentation of the hybrid skeleton comprises: segmenting the hybrid skeleton; and obtaining the segmentation of the three-dimensional shape by the segmentation of the mixed type framework according to the corresponding relation between the mixed type framework and the sampling points.
Wherein said obtaining a substructure of the three-dimensional shape from the segmented three-dimensional shape comprises: obtaining a plurality of parts of the three-dimensional shape from the divided three-dimensional shape; creating a connection diagram connecting the plurality of components; and extracting subgraphs in the connection graph as substructures of the three-dimensional shape.
Wherein said building a representation of said three-dimensional shape using a bag-of-words model from a substructure of said three-dimensional shape comprises: matching the sub-structure of the three-dimensional shape with each candidate sub-structure in a set of candidate sub-structures to determine a frequency of occurrence of the each candidate sub-structure on the three-dimensional shape; creating a word vector for the three-dimensional shape according to the frequency of occurrence of the respective candidate substructures in the three-dimensional shape; and normalizing the word vectors to obtain the word bag expression of the three-dimensional shape.
Wherein before the using the bag-of-words model to build the expression of the three-dimensional shape, the method further comprises: creating the set of candidate substructures; wherein the creating the set of candidate substructures comprises: obtaining all three-dimensional shape substructures in an input dataset; determining similarity between the obtained substructures; selecting the candidate substructures from the acquired substructures according to similarities between the acquired substructures to form the set of candidate substructures.
Wherein the determining the similarity between the obtained substructures comprises: defining a graph kernel between the acquired substructures; determining similarities between the acquired substructures from the graph kernels.
Wherein the defining a graph kernel between the obtained substructures comprises: defining a node core and an edge core; wherein the node core is:
the edge core is as follows:
k isnode(ni,nj) Represents a node core, kedge(ei,ej) Denotes the edge nucleus, ni,njRepresents a node, hiAnd hjAre respectively related to the node niAnd njIs connected by a histogram of the geometric features of the components (D) is hiAnd hjThe normalized correlation of (a) is calculated,
are any two pairs hiAnd hjDistance D (h) ofi,hj) Maximum value of uiAnd ujIs a two-dimensional histogram of all point pairs in the two connected components with respect to vertical upward angle and distance.
The embodiment of the invention provides a three-dimensional shape expression device, which comprises: the framework extraction module is used for extracting a three-dimensional mixed framework; the segmentation module is used for obtaining the segmentation of the three-dimensional shape by segmenting the mixed type skeleton; the substructure extraction module is used for obtaining a substructure of the three-dimensional shape according to the segmented three-dimensional shape; and the expression module is used for establishing the expression of the three-dimensional shape by utilizing a bag-of-words model according to the substructure of the three-dimensional shape.
Wherein, the skeleton extraction module is specifically configured to: sampling the surface of the three-dimensional shape to obtain a sampling point; and re-expressing the sampling points to obtain a mixed skeleton containing a one-dimensional curve and a two-dimensional slice.
The segmentation module is specifically used for segmenting the hybrid skeleton; and obtaining the segmentation of the three-dimensional shape by the segmentation of the mixed type framework according to the corresponding relation between the mixed type framework and the sampling points.
The embodiment of the invention has the beneficial effects that:
according to the embodiment of the invention, word bag expression based on the substructure is established for the three-dimensional shape, and the expression mode has the characteristics of simplicity, high efficiency and the like.
Drawings
FIG. 1 is a schematic flow chart diagram of an embodiment of a three-dimensional shape representation method of an embodiment of the present invention;
FIG. 2 is a schematic flow chart of an embodiment of step 101 in FIG. 1;
FIG. 3 is a schematic flow chart diagram of an embodiment of step 102 of FIG. 1;
FIG. 4 is a schematic flow chart of an embodiment of step 103 in FIG. 1;
FIG. 5 is a schematic flow chart diagram of an embodiment of step 104 in FIG. 1;
FIG. 6 is a schematic illustration of a three-dimensionally shaped hybrid skeleton;
FIGS. 7(a) - (c) are schematic views of three partial features for hybrid skeleton segmentation, respectively;
FIG. 8 is a schematic diagram of a process for creating a three-dimensional shape graph structure;
FIG. 9 is a schematic representation of a bag of words in a three-dimensional shape;
FIG. 10 is a schematic diagram of performing a three-dimensional shape search;
fig. 11 is a schematic structural diagram of an embodiment of a three-dimensional shape expression apparatus according to an embodiment of the present invention.
Detailed Description
In order to make the technical problems, technical solutions and advantageous effects solved by the present invention more clearly apparent, the present invention is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
As shown in fig. 1, it is a schematic flow chart of an embodiment of a three-dimensional shape expression method according to an embodiment of the present invention, and the method includes the following steps:
step 101: and extracting the three-dimensional mixed skeleton.
Step 102: the three-dimensional shape is obtained by dividing the hybrid skeleton.
Step 103: and obtaining a substructure of the three-dimensional shape according to the divided three-dimensional shape. And
step 104: and establishing the expression of the three-dimensional shape by utilizing the bag-of-words model according to the substructure of the three-dimensional shape.
Wherein, the hybrid skeleton of step 101 includes: one-dimensional curves and two-dimensional flakes. Fig. 2 is a schematic flow chart of an embodiment of step 101. Specifically, in order to extract the hybrid skeleton, first, a surface of a three-dimensional shape is sampled to obtain a plurality of sampling points (step 201); then, all the sampling points are re-expressed to obtain a mixed skeleton containing a one-dimensional curve and a two-dimensional slice (step 202). In step 201, a poisson disk sampling method may be used to sample the three-dimensional surface, where the number of sampling points is set to about 20000. The sampling points obtained in step 201 are surface points, and in step 202, the sampling points can be expanded to be depth points, each surface point is combined with a corresponding skeleton point inside the three-dimensional shape, the direction of the depth point connecting line is consistent with the normal vector of the surface point by optimizing the arrangement of the points on the surface and the skeleton of the shape, and finally the convergence of the optimization function obtains a mixed skeleton consisting of a one-dimensional curve and a two-dimensional slice. For example, as shown in fig. 6, a schematic view of a three-dimensional hybrid skeleton of a chair is shown.
Fig. 3 is a schematic flow chart of an embodiment of step 102. Specifically, first, the hybrid skeleton is divided (step 301): then, according to the corresponding relationship between the hybrid skeleton and the sampling points, the three-dimensional shape segmentation can be obtained by the hybrid skeleton segmentation (step 302). In step 301, three local features based on PCA (Principal Component Analysis) are calculated for each hybrid skeleton point in the hybrid skeleton. Calculating the three characteristics of each point requires selecting a geodesic neighborhood of the point and calculating the characteristic value lambda of the point1≥λ2≥λ3≧ 0, and define:
l, P, S is the above three local features, which respectively describe the degree of linearity, planarity, sphericity of the shape of the point neighborhood; for example, as shown in fig. 7(a) - (c), they are schematic diagrams of the above three local features, that is, fig. 7(a) - (c) respectively show the linearity, planarity and sphericity of the region in a visualized manner. Wherein, the local features are used for the clustering process of segmenting the mixed skeleton. In addition, in the embodiment of the invention, in order to obtain a more semantic three-dimensional shape segmentation, a semi-supervised spectral clustering method can be adopted, and the manual interaction of the mixed type skeleton is combined to obtain the segmentation of the mixed type skeleton. Wherein, the process of manual interaction allows two constraints of 'must connect' and 'cannot connect' to be set for different points on the mixed type skeleton to guide the segmentation result.
As shown in fig. 4, is a schematic flow chart of an embodiment of step 103. Specifically, in fig. 4, first, from the divided three-dimensional shape, a plurality of parts of the three-dimensional shape are obtained (step 401); then, a connection map connecting the plurality of components is created (step 402); finally, the subgraph in the connected graph is extracted as the substructure of the three-dimensional shape (step 403). The nodes of the connection diagram obtained in step 402 are the components in the three-dimensional shape, and when the connection diagram is created, if the distance between any point pair in any two components is less than 2% of the length of the diagonal line of the whole three-dimensional shape bounding box, the two components are connected by one edge. In step 403, a sub-graph with the number of connected nodes n being 1, …,5 in the connection graph is extracted, so as to obtain a sub-structure corresponding to the three-dimensional shape, where the sub-structure may be represented by a series of geometric features, where the value of n is not limited to the above example. As shown in fig. 8, a process of creating a three-dimensional shape diagram structure is shown, where the left shape in fig. 8 is a schematic diagram of segmentation for obtaining a hybrid skeleton after human interaction is combined, the middle shape in fig. 8 is a segmentation for obtaining a three-dimensional shape through the association of the hybrid skeleton and a shape surface point, and the right shape in fig. 8 is a diagram for establishing connection between components of the three-dimensional shape.
Fig. 5 is a schematic flow chart of an embodiment of step 104. Specifically, first, a sub-structure of the three-dimensional shape is matched with each candidate sub-structure in the candidate sub-structure set to determine the frequency of occurrence of each candidate sub-structure on the three-dimensional shape (step 501); then, creating a word vector of the three-dimensional shape according to the frequency of occurrence of each candidate substructure in the three-dimensional shape (step 502); finally, the word vectors are normalized to obtain a representation of the three-dimensional shape (step 503).
Therein, the set of candidate substructures in step 501 may be created in advance, for example, by information of all three-dimensional shapes in the input data set. Specifically, all three-dimensional shape substructures in the input dataset may be obtained, and the obtaining manner herein may include, for example: respectively extracting and segmenting the mixed skeleton of each three-dimensional shape to obtain the segmentation of the three-dimensional shape, thereby obtaining the substructure of the three-dimensional shape; then determining the similarity between the obtained substructures; and finally, selecting the candidate substructures from the acquired substructures according to the similarity between the acquired substructures to form a candidate substructure set. In addition, when selecting the candidate substructures, the number of candidate substructures may also be determined at the same time.
Wherein, in order to select a representative substructure, the distance between two substructures is used. And the distance between two substructures with the same number of components is calculated, and the similarity of two subgraphs with the same number of nodes can be obtained through graph kernel calculation, so that the similarity between the substructures is obtained. In the calculation process, a node kernel (node kernel) and an edge kernel (edge kernel) are used, and in this embodiment, the node kernel is defined as:
wherein h isiAnd hjAre respectively related to node niAnd njThe histogram of all geometric features of the component(s) of (a) is connected, the geometric features including the Shape Diameter Function (Shape Diameter Function) and the three PCA-based local features previously used, and the dimension of each feature histogram is 16. D is hiAnd hjNormalized correlation (normalized correlation). Symbol
Are any two pairs hiAnd hjDistance D (h) ofi,hj) Is measured.
Wherein the edge kernel is used to capture the similarity of two pairs of connected components, which is defined as:
wherein u isiAnd ujIs a two-dimensional histogram of all point pairs in the two connected components with respect to vertical upward angle and distance. The two characteristics are obtained by calculating the distance of a line segment formed by the point pairs and the included angle formed by the line segment and the vertical upward direction of the three-dimensional shape. The similarity of the two substructures is obtained by adding and summing the similarities of graph steps (graph walks) less than or equal to p in the graph core, wherein p is the number of the nodes of the substructures. That is, the similarity between two substructures is calculated by a graph walk kernel (graph walk kernel) considering the node kernel and the edge kernel.
After the similarity of the substructures is defined, a candidate substructures set C can be extracted from all the initial substructures (i.e., the substructures of all the three-dimensional shapes described above) to build a dictionary in the bag-of-words model. One problem to be solved for the initial set of substructures is that there are a large number of similar substructures. The main reason for this problem is that the three-dimensional shape substructure is not very distinctive. Therefore, in order to avoid processing a large number of similar or unrelated substructures, the invention selects and obtains a candidate substructure set by performing density analysis in the substructure similarity space. I.e. calculating the density of all sub-structures and then only retaining those sub-structures whose density is at the peak. The density peak is associated with the cluster center of the initial set of substructures having the same number of nodes. Therefore, only the sub-structures surrounded by the similar sub-structures in the similar space and having the peak density are selected, so that the redundant processing of the similar sub-structures can be avoided. Also, since these substructures are density peaks, they frequently occur in the set of substructures.
In order to compute the density in a relatively robust manner, a clustering approach may be employed. Before clustering, it is assumed that the cluster center is surrounded by nearby substructures of lower density values and is preserved with other substructures of higher density valuesAt a relatively large distance. Specifically, first, the distance between the two substructures is calculated from the definition of the similarity between the graph kernel (graph kernel) and the above-mentioned substructures, and is denoted as dij. Then, a certain substructure eiLocal density of (p)iIs defined as:
ρi＝∑jχ(dij-dc)，
wherein, when x<1, x (x) is 1, otherwise 0; dcFor the cutoff distance (cutoff distance), d is set herecThe 2 nd% values after sorting all the inter-substructure distance values from small to large. Next, the distance δ from one substructure to another higher density substructureiIs defined as:
and the highest density substructure as a special case, its distance to other high density substructures will be defined as δi＝maxjdij. Finally, the values of delta are assigned to all substructuresiSorting and selecting the largest first K substructures to obtain the maximum distance value deltai. The number of substructures K can be set empirically, for example, according to the variance γi＝ρiδiTo decide. In the above manner, a candidate substructure set C is finally obtained. It is noted that the above method is handled separately for sub-structures containing different numbers of components.
After the candidate substructure set C is obtained, the frequency of occurrence of the candidate substructure set C on each three-dimensional shape is represented by a vector t according to the concept of the bag-of-words model, which is called a word vector (term vector) of the three-dimensional shape. t is an m-dimensional vector, where m is the number of substructures in the set of candidate substructures C. Each dimension in t is counted the number of times a candidate sub-structure appears on the three-dimensional shape, and then normalized by the number of all sub-structures on the three-dimensional shape. If a related substructure does not appear on the three-dimensional shape, the corresponding dimension value is zero.
To create a word vector for a given three-dimensional shape S, a similar procedure is used to extract the substructure from the three-dimensional shape as when the initial substructure was obtained. Then, only the substructures associated with the candidate substructure set C are retained, and the detected substructures are counted to set the values corresponding to the associated substructures in the word vector. In order to find similar substructures in the three-dimensional shape S as in the candidate set of substructures C, the similarity between the substructures needs to be considered. For the substructures in the graph, given a substructure S epsilon S and a candidate substructure C epsilon C on the three-dimensional shape, if the kernel distance (kernel distance) between the substructure S and the substructure C is less than the threshold value tausThen the two substructures are considered similar. In order to obtain the threshold value for each candidate substructure c, the kernel distance between the substructure c and other substructures needs to be calculated first, and a histogram needs to be established. The histogram is then fitted to the Beta distribution and τ is setsThe value is where the inverse cumulative distribution function value is 0.05, which means that 95% of the distances of other substructures to that substructure are greater than τs。
As shown in fig. 9, a three-dimensional chair shape and a three-dimensional stool shape are illustrated, a and b represent word bag expressions, i.e., word vectors, of the two three-dimensional shapes, respectively, and the word vector t of each three-dimensional shape is represented by an m-dimensional histogram. The same position in the histogram corresponds to a common candidate substructure. In fig. 9, it can be seen that both three-dimensional shapes have a common "cushion and support leg" substructure. For stool shapes, which are of a different shape type than chair shapes, and do not have the "arm" substructure in a chair, there is no corresponding data in the corresponding dimension of the histogram.
In the existing mode, when a three-dimensional shape is expressed, the requirement on data input is high; the embodiment of the invention has strong adaptability to the input three-dimensional shape. Meanwhile, the three-dimensional shape expression of the invention is based on the expression of the substructure, so that the global framework of the three-dimensional shape can be captured better, and the three-dimensional shape data with the problems of noise and loss can be adapted. In addition, the conventional three-dimensional shape expression uses two-dimensional pictures, but usually needs a plurality of angles to completely capture the form of the three-dimensional shape, so that a large number of two-dimensional pictures are used; the word bag expression utilized by the invention is only a simple histogram finally, and can well reflect the structure of the three-dimensional shape, and the application and the expansion of the invention are more convenient. In addition, in the invention, when the final expression result of the three-dimensional shape is used for comparison, search and the like, the distance can be directly calculated, and further processing and transformation are not needed on the basis of the distance.
The foregoing illustrates the principles of embodiments of the present invention and the following examples illustrate the application of embodiments of the present invention.
The three-dimensional shape expression mode of the embodiment of the invention can be applied to the fields of three-dimensional shape retrieval, comparison, classification, identification and the like. The following explains an application of an embodiment of the present invention by taking a search as an example. For a given newly input three-dimensional shape, segmenting and extracting an initial substructure in the three-dimensional shape, and then matching candidate substructures according to the similarity among the substructures to obtain the occurrence frequency of all candidate substructures, thereby completing the process of expressing the three-dimensional shape by using a bag-of-words model. Then, calculating a distance value between the newly input three-dimensional shape expression and each three-dimensional shape expression in the data set, and obtaining a retrieval result according to the calculated distance value; for example, the smaller the distance, the more similar the two three-dimensional shapes are represented. For example, as shown in FIG. 10, a diagram of the results of a three-dimensional shape retrieval application is shown, where the left-hand box is the three-dimensional shape input for the retrieval process, and the right-hand box is the result returned by the retrieval. In fig. 10, the first ten corresponding three-dimensional shapes of the minimum distance are taken as the returned search results, and as a result, their outer configuration structures are also most similar to the searched three-dimensional shapes.
Method examples of embodiments of the present invention are described above, and apparatus examples of embodiments of the present invention are described below.
Fig. 11 is a schematic structural view of an embodiment of the three-dimensional shape expression device according to the present invention. It includes: a skeleton extraction module 111 for extracting a three-dimensional shaped hybrid skeleton; a segmentation module 112, configured to obtain a segmentation of the three-dimensional shape by segmenting the hybrid skeleton; a substructure extraction module 113, configured to obtain a substructure of the three-dimensional shape according to the segmented three-dimensional shape; and an expression module 114 for building an expression of the three-dimensional shape using a bag-of-words model according to the substructure of the three-dimensional shape. Wherein, the skeleton extraction module 111 is specifically configured to: sampling the surface of the three-dimensional shape to obtain a sampling point; and re-expressing the sampling points to obtain a mixed skeleton containing a one-dimensional curve and a two-dimensional slice. The segmentation module 112 is specifically configured to segment the hybrid skeleton; and obtaining the segmentation of the three-dimensional shape by the segmentation of the mixed type framework according to the corresponding relation between the mixed type framework and the sampling points.
It should be noted that the functions and actions of the modules in the apparatus correspond to steps 101 to 104 in the above method embodiment, respectively, and since the above steps are described in detail in the foregoing, the description is not repeated for brevity.
It will be understood by those skilled in the art that all or part of the processes of the methods of the embodiments described above can be implemented by a computer program, which can be stored in a computer-readable storage medium, and when executed, can include the processes of the embodiments of the methods described above. The storage medium may be a magnetic disk, an optical disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), or the like.
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents and improvements made within the spirit and principle of the present invention are intended to be included within the scope of the present invention.
Claims (6)
1. A three-dimensional shape expression method, comprising the steps of:
extracting a three-dimensional shape mixed skeleton;
obtaining the segmentation of the three-dimensional shape by segmenting the mixed type skeleton;
obtaining a substructure of the three-dimensional shape according to the segmented three-dimensional shape; and
establishing a representation of the three-dimensional shape using a bag-of-words model according to the substructure of the three-dimensional shape;
the hybrid skeleton for extracting the three-dimensional shape comprises:
sampling the surface of the three-dimensional shape to obtain a sampling point; and
re-expressing the sampling points to obtain a mixed skeleton containing a one-dimensional curve and a two-dimensional slice;
the obtaining of the segmentation of the three-dimensional shape by the segmentation of the hybrid skeleton includes:
obtaining the segmentation of the mixed type framework by adopting a semi-supervised spectral clustering method and combining with manual interaction on the mixed type framework;
according to the corresponding relation between the mixed type framework and the sampling points, the three-dimensional shape is obtained by the division of the mixed type framework;
said obtaining a substructure of said three-dimensional shape from said segmented three-dimensional shape comprising:
obtaining a plurality of parts of the three-dimensional shape from the divided three-dimensional shape;
creating a connection diagram connecting the plurality of components;
and extracting subgraphs in the connection graph as substructures of the three-dimensional shape.
2. A method of expressing a three-dimensional shape according to claim 1, wherein said using a bag-of-words model to build an expression of said three-dimensional shape from a substructure of said three-dimensional shape comprises:
matching the sub-structure of the three-dimensional shape with each candidate sub-structure in a set of candidate sub-structures to determine a frequency of occurrence of the each candidate sub-structure on the three-dimensional shape;
creating a word vector for the three-dimensional shape according to the frequency of occurrence of the respective candidate substructures in the three-dimensional shape;
and normalizing the word vectors to obtain the word bag expression of the three-dimensional shape.
3. A method of expressing a three-dimensional shape as defined in claim 2, wherein prior to using the bag-of-words model to build the expression of the three-dimensional shape, further comprising:
creating the set of candidate substructures;
wherein the creating the set of candidate substructures comprises:
obtaining all three-dimensional shape substructures in an input dataset;
determining similarity between the obtained substructures;
selecting the candidate substructures from the acquired substructures according to similarities between the acquired substructures, thereby forming the set of candidate substructures.
4. The method of claim 3, wherein said determining similarity between said acquired substructures comprises:
defining a graph kernel between the acquired substructures;
determining similarities between the acquired substructures from the graph kernels.
5. The method of claim 4, wherein the defining a kernel between the acquired substructures comprises:
defining a node core and an edge core;
wherein the node core is:
the edge core is as follows:
k isnode(ni,nj) Represents a node core, kedge(ei,ej) Denotes the edge nucleus, ni,njRepresents a node, hiAnd hjAre respectively related to the node niAnd njA histogram formed by connecting the geometric feature histograms of the components of (a), D (h)i,hj) Is hiAnd hjThe normalized correlation distance of (a) is,
are any two pairs hiAnd hjDistance D (h) ofi,hj) Is measured.
6. A three-dimensional shape expressing apparatus using the three-dimensional shape expressing method according to any one of claims 1 to 5, comprising:
the skeleton extraction module is used for extracting a three-dimensional shape hybrid skeleton, and the process of extracting the three-dimensional shape hybrid skeleton comprises the following steps: sampling the surface of the three-dimensional shape to obtain a sampling point, and re-expressing the sampling point to obtain a mixed skeleton containing a one-dimensional curve and a two-dimensional slice;
the segmentation module is used for segmenting the mixed type framework and obtaining the segmentation of the three-dimensional shape according to the corresponding relation between the mixed type framework and the sampling points;
the substructure extraction module is used for obtaining a substructure of the three-dimensional shape according to the segmented three-dimensional shape; and
and the expression module is used for establishing the expression of the three-dimensional shape by utilizing a bag-of-words model according to the substructure of the three-dimensional shape.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710171841.2A CN106960032B (en) | 2017-03-21 | 2017-03-21 | Three-dimensional shape expression method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710171841.2A CN106960032B (en) | 2017-03-21 | 2017-03-21 | Three-dimensional shape expression method and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106960032A CN106960032A (en) | 2017-07-18 |
| CN106960032B true CN106960032B (en) | 2021-02-19 |
Family
ID=59471025
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710171841.2A Active CN106960032B (en) | 2017-03-21 | 2017-03-21 | Three-dimensional shape expression method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106960032B (en) |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101308500A (en) * | 2008-05-23 | 2008-11-19 | 浙江大学 | Visual and efficient three-dimensional human body movement data retrieval method based on demonstrated performance |
| CN101488142A (en) * | 2008-12-09 | 2009-07-22 | 南京大学 | Three-dimensional solid model retrieval method based on face topological interconnection constraint |
| CN101763652A (en) * | 2009-06-03 | 2010-06-30 | 中国科学院自动化研究所 | Three-dimensional framework fast extraction method based on branch feathers |
| CN106021330A (en) * | 2016-05-06 | 2016-10-12 | 浙江工业大学 | A three-dimensional model retrieval method used for mixed contour line views |
-
2017
- 2017-03-21 CN CN201710171841.2A patent/CN106960032B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101308500A (en) * | 2008-05-23 | 2008-11-19 | 浙江大学 | Visual and efficient three-dimensional human body movement data retrieval method based on demonstrated performance |
| CN101488142A (en) * | 2008-12-09 | 2009-07-22 | 南京大学 | Three-dimensional solid model retrieval method based on face topological interconnection constraint |
| CN101763652A (en) * | 2009-06-03 | 2010-06-30 | 中国科学院自动化研究所 | Three-dimensional framework fast extraction method based on branch feathers |
| CN106021330A (en) * | 2016-05-06 | 2016-10-12 | 浙江工业大学 | A three-dimensional model retrieval method used for mixed contour line views |
Non-Patent Citations (1)
| Title |
|---|
| 拓扑和形状特征相结合的三维模型检索;王飞等;《计算机辅助设计与图形学学报》;20080409;正文第[2]段、第[15]-[17]段及图2-3 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106960032A (en) | 2017-07-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106570505B (en) | Method and system for analyzing histopathological images | |
| Unnikrishnan et al. | Toward objective evaluation of image segmentation algorithms | |
| US8582887B2 (en) | Image processing system, learning device and method, and program | |
| Lucchi et al. | Learning for structured prediction using approximate subgradient descent with working sets | |
| WO2022041678A1 (en) | Remote sensing image feature extraction method employing tensor collaborative graph-based discriminant analysis | |
| Panagiotakis et al. | Natural image segmentation based on tree equipartition, bayesian flooding and region merging | |
| US11301509B2 (en) | Image search system, image search method, and program | |
| Xu et al. | Classification of LiDAR point clouds using supervoxel-based detrended feature and perception-weighted graphical model | |
| CN109086777B (en) | Saliency map refining method based on global pixel characteristics | |
| CN110188763B (en) | Image significance detection method based on improved graph model | |
| KR101977231B1 (en) | Community detection method and community detection framework apparatus | |
| Srivastava et al. | Looking for shapes in two-dimensional cluttered point clouds | |
| Mohamed et al. | Deformable 3d shape retrieval using a spectral geometric descriptor | |
| Tu et al. | Feature extraction using multitask superpixel auxiliary learning for hyperspectral classification | |
| Wang et al. | Cluttered indoor scene modeling via functional part-guided graph matching | |
| CN116778339A (en) | Method and system for selecting hyperspectral wave bands by aid of local view auxiliary discrimination | |
| CN106960032B (en) | Three-dimensional shape expression method and device | |
| WO2018170731A1 (en) | Method and device for expressing three-dimensional shape | |
| US9953458B2 (en) | Method and apparatus for generating shape descriptor of a model | |
| Li et al. | Feature proposal model on multidimensional data clustering and its application | |
| Bu et al. | Shift-invariant ring feature for 3D shape | |
| Khelifi et al. | Semantic image segmentation using the ICM algorithm | |
| CN112884884A (en) | Candidate region generation method and system | |
| Wei et al. | Multi-task joint learning of 3d keypoint saliency and correspondence estimation | |
| Bhattacharya et al. | DT-RANSAC: a delaunay triangulation based scheme for improved RANSAC feature matching |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| EE01 | Entry into force of recordation of patent licensing contract |
Application publication date: 20170718 Assignee: Shenzhen achievement transformation Center Co.,Ltd. Assignor: SHENZHEN INSTITUTES OF ADVANCED TECHNOLOGY CHINESE ACADEMY OF SCIENCES Contract record no.: X2023980044064 Denomination of invention: Method and device for expressing three-dimensional shapes Granted publication date: 20210219 License type: Common License Record date: 20231023 |
|
| EE01 | Entry into force of recordation of patent licensing contract |