CN101916284B - Three-dimensional model searching method based on shape orientation multi-resolution analysis - Google Patents
Three-dimensional model searching method based on shape orientation multi-resolution analysis Download PDFInfo
- Publication number
- CN101916284B CN101916284B CN2010102582422A CN201010258242A CN101916284B CN 101916284 B CN101916284 B CN 101916284B CN 2010102582422 A CN2010102582422 A CN 2010102582422A CN 201010258242 A CN201010258242 A CN 201010258242A CN 101916284 B CN101916284 B CN 101916284B
- Authority
- CN
- China
- Prior art keywords
- sampling
- function
- shape
- sample plane
- towards
- 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.)
- Expired - Fee Related
Links
Images
Abstract
The invention discloses a three-dimensional model searching method based on shape orientation multi-resolution analysis, and the method comprises the following steps: obtaining an orientation sampling plane by adopting a three-dimensional plane patch orientation principle component analysis method; sampling orientation distribution of the three-dimensional shape and generating a sampling function of the orientation distribution; analyzing the sampling function and extracting a multi-resolution low-frequency wavelet coefficient of the sampling function; and calculating feature distance in a self-adaption multi-resolution feature distance calculating method. The method can enhance the precision and speed of searching the three-dimensional model and can be applied to the fields of three-dimensional model multiplex, model identification, robot vision, virtual reality and the like.
Description
Technical field
The present invention relates to a kind of method for searching three-dimension model.
Background technology
In the past two during the last ten years, along with the widespread use of lifting, multi-media recording and the capture apparatus of the popularizing of computing machine, hardware performance, multimedia messages develops rapidly in every field, lets All Around The World become very abundant and colorful.Multimedia messages mainly comprises sound, image, video, three-dimensional model.At present on the internet, be full of the multimedia messages of magnanimity on individual and the industrial user's hard disk, caused information in each media explosive increase.How to help the user in huge volumes of content, to find needed multimedia messages, become the research object of multimedia coupling and retrieval technique.The design of each part media work and make and all to need very big effort with for a long time, the designer can obtain analog information through multimedia retrieval and Information Reuse, and new data of modification making a little, increases work efficiency with this; Robot vision is retrieved recognition objects such as subject image in the robot memory bank through the multimedia coupling; Fields such as strick precaution, checking, detection utilize the multimedia matching technique to come coupling relevant informations such as query fingerprints storehouse, two-dimensional/three-dimensional face database, three-dimensional head storehouse; Industry spot can be judged control information, fault type etc. automatically according to images match; Field of traffic carries out record violating the regulations automatically, the charge of high speed crossing etc. through car plate identification etc.Yet the coupling of content of multimedia and retrieval do not have content of text coupling and retrieval such simple; From one piece of literal, extract crucial literal information; Can represent the descriptor of this image to compare with from multimedias such as piece image, extracting, the mankind that the latter only has senior thought just can realize fully.Extracting multimedia characteristic through computing machine is a very complicated job, is multimedia coupling and the core technology of retrieving.
Three-dimensional digital model be a kind of than two dimensional image content more horn of plenty meet human visual system truly, more, more can clearly express the multimedia data type of a real three-dimensional object; Its basic data is the 3-dimensional digital shape, realizes three-dimensional digital model by shape.In the last few years; Wideling popularize of D modeling tool such as novel 3D shape obtains that equipment is brought in constant renewal in, the progress of 3D shape Modeling Research and MAYA, 3DMAX, Pro/E, CATIA make the design of 3D shape become more and more simpler, three-dimensional model in the multimedia field, aspect widespread uses such as virtual reality, industrial design and manufacturing CAD/CAM, MRI medical 3 D image, biomolecule and gene structure, recreation, cartoon role, man-machine interaction.These widespread uses make has the three-dimensional model of counting in million to produce every day, under the Internet technology effect, makes bamboo telegraph simultaneously, exists three-dimensional model is mated and the active demand of retrieving.And; Three-dimensional model coupling and retrieval technique can be applied in during three-dimensional model reuses; Because three-dimensional model needs a large amount of modeling work, it can help the designer to find scale model rapidly and on existing model basis, construct new model, practices thrift the modeling time.In addition, can also be applied in fields such as multimedia information retrieval, cad model coupling, robot three-dimensional vision, biomolecule and medical science organ model coupling, mechanical model designing support, three dimensional object identification and checking, virtual reality scenario object matches, animation and game role designing support, the moon and deep space object-detection, fault diagnosis.Content-based 3D shape coupling studied with retrieval technique become extremely important.
At present, the efficient content-based fast three-dimensional model of realization matees a research focus that has become domestic and international multimedia messages identification and searching field with retrieval technique.In the domestic and international disclosed document, at T.Funkhouser, P.Min, M.Kazhdan; J.Chen, A.Halderman, D.Dobkin, and D.Jacobs; " A search engine for 3Dmodels, " ACM Transactions on Graphics, Vol.22, No.1; Pp.83-105 proposed the employing spherical harmonic function in 2003. and analyzed the voxelization model, extracted the characteristic that harmonic coefficient is described 3D shape; H.Laga, H.Takahashi, and M.Nakajima; " Spherical wavelet descriptors for content-based 3D modelretrieval, " IEEE International Conference on Shape Modeling and Applications, Matsushima; Japan; Pp.15-25, employing spherical wavelet coefficient analysis has been proposed among the Jun.2006. sphere prolong function, and adopt wavelet coefficient to describe the characteristic of 3D shape; T.Furuya and R.Ohbuchi; " DenseSampling and Fast Encoding for 3D Model Retrieval Using Bag-of-Visual Features; " ACMInternational Conference on Image and Video Retrieval, Santorini, Greece; Pp.1-8 has proposed the proper vector dimension reduction method of depth buffer figure among the Jul.2009..
But the feature extracting method that above-mentioned document proposes has some not enough:
(1) above method is extracted shape facility through computed range.Yet, calculate the depth buffer distance of shape face to the distance of barycenter or shape face to bounding box, if barycenter or bounding box generation subtle change, great changes will take place can to cause the distance function of extraction, the poor stability of method;
(2) the shape criteria process of some method has adopted principal component analysis (PCA) PCA or continuous principal component analysis (PCA) CPCA method, because the main shaft stability of calculating based on PCA and CPCA method is not high enough, impacts for follow-up feature extracting method;
(3) certain methods is considered lessly on the time complexity of feature extraction algorithm, and the time of make extracting characteristic is longer, has lost feature extraction precision and the balance of feature extraction time, has influenced real-time online application.
Summary of the invention
In order to overcome the long and low deficiency of characteristic area component of prior art poor stability, extraction time; The present invention provides a kind of three-dimensional shape features method for distilling based on shape orientation multi-resolution analysis; This method has increased the stability of method; Guarantee that pre-service to the feature extraction in the later stage little influence of trying one's best, has improved the speed of feature extraction, strengthened the differentiation power of characteristic.
The technical solution adopted for the present invention to solve the technical problems may further comprise the steps:
(1) adopt the dough sheet of 3D shape to obtain towards sample plane towards principal component analytical method: the dough sheet that adopts 3D shape towards principal component analytical method obtain three towards main shaft; And six shape orientation sample plane are set according to these three vertical major; This sample plane spatially constitutes sealing, in 3D shape is enclosed in.This step need each dough sheet of Calculation of Three Dimensional shape towards the method for average vector of normal vector and 3D shape; The normal vector covariance matrix that constitutes thus; In this matrix; The normal vector of each dough sheet is carried out the weighting of area size, the eigenwert of rank order matrix and corresponding three proper vectors by size, thereby according to three proper vectors confirming as towards main shaft to calculate the sample plane of shape orientation.Sample plane satisfies two conditions: (a) each proper vector is confirmed two sample plane; This proper vector is vertical with two sample plane; And these two sample plane should lay respectively at the both sides of three-dimensional model, do not intersect with three-dimensional model, and three proper vectors are determined six sample plane altogether; (b) in six sample plane are fully enclosed in three-dimensional model and since sample plane will gather 3D shape towards parameter, so sample plane can set arbitrarily apart from the distance of three-dimensional model, this notion with bounding box is different.
(2) sampling 3D shape towards distribution; Generation is towards the sampling function that distributes: the sample plane through step (1) generates is evenly launched the sampling ray according to some levels; Intersect with the surface of 3D shape, the inner product of calculating sampling ray and crossing with it surface normal is with the functional value of inner product value as the shape orientation function; The index of employing 2 doubly carries out uniform sampling during sampling, is beneficial to adopt fast method that sampling function is analyzed.
(3) analytical sampling function; Extract the multi-resolution low-frequency wavelet coefficient of sampling function: select the advantages of simplicity and high efficiency wavelet mother function, owing to be function of region towards function, what relatively be good at for region-operation is generating function such as Daubechies; Displacement and change of scale through this generating function; Can constitute one group of function base, utilize function base and according to scaling function and wavelet function to decomposing according to yardstick towards function, the method that what adopt for two-dimentional sampling function is advanced every trade conversion, carry out rank transformation is again carried out dimensionality reduction; Obtain the many groups low frequency wavelet coefficient under a plurality of resolution, the low frequency wavelet coefficient constitutes the characteristic of multiresolution.
(4) through self-adaptation multiresolution features distance calculating method calculated characteristics distance: utilize the characteristic distance under certain resolution between two three-dimensional models of minimum Manhattan distance calculation of many sampling functions, utilize the characteristic distance that the resolution method of weighting is calculated following two three-dimensional models of all resolution simultaneously.The method of weighting that is adopted is adaptive weighted, utilizes sample database to make the First Tier parameter maximization in the search theory, to obtain optimum multiresolution features weighting coefficient.Make characteristic distance maximization and optimization through this apart from balancing method.
The invention has the beneficial effects as follows: the present invention considered 3D shape towards, and abandoned extracting the traditional approach of distance feature, thereby increased the stability of method.Extracting preceding pretreatment stage to three-dimensional shape features thinks better of and proves; Discovery has certain influence based on the preprocess method of principal component analysis (PCA) PCA and continuous principal component analysis (PCA) CPCA to follow-up feature extraction; Therefore the present invention proposes and a kind ofly obtain the shape sample plane based on the shape orientation principal component analytical method; Because the major component of shape orientation is relatively stable; Compare with the preprocessing process of common principal component analysis (PCA) PCA or continuous principal component analysis (PCA) CPCA, less to later stage feature extraction influence.When obtaining primitive character; The present invention proposes shape orientation as the most original analytic target; Compare with method before, in the robustness of strengthening feature extraction, can under shape orientation sampling rate seldom, obtain number of characteristics based on apart from primitive character.It has adopted the multiresolution wavelet analytical approach that the shape orientation function has been carried out express-analysis, has improved the speed that characteristics of low-frequency is extracted, and helps method is pushed to the application scenario of real-time online.The present invention proposes self-adaptation multiresolution features distance calculating method, have different distance to weigh mode,, improved characteristic area component to guarantee the maximization and the optimization of characteristic distance to guarantee different resolution.
Below in conjunction with accompanying drawing and embodiment the present invention is further specified.
Description of drawings
Fig. 1 is the system flowchart of three-dimensional model search according to the invention;
Fig. 2 is the process flow diagram of three-dimensional model feature extraction of the present invention;
Fig. 3 is the key diagram of the shape orientation method of sampling of the present invention;
Fig. 4 is the accelerated method key diagram of the shape orientation method of sampling of the present invention;
Fig. 5 is the key diagram that obtains towards the multiresolution coefficient of sampling function of the present invention.
Embodiment
Detailed explanation as shown in Figure 1 the system flowchart of three-dimensional model search.This technology of the present invention can realize the systemic-function of three-dimensional model web search.Import a three-dimensional model through the user at query interface; All scale models of this model of plan search; Searching system sends the retrieval request of this model to search engine through network; Search engine is made response, will retrieve qualified three-dimensional model and send the user to, and according to the similarity height three-dimensional model sorted.Idiographic flow is; Receive the retrieval request of the three-dimensional model that the client user sends in system; At first obtain the proper vector of three-dimensional model through feature extraction algorithm; This proper vector can be unique three-dimensional model of expression and also have the height ability to see things in their true light, can distinguish difference with other three-dimensional models.Feature extraction algorithm is analyzed shape orientation, obtains multiresolution features.System adopts the self-adaptive features comparative approach to calculate one by one the distance of this proper vector and property data base, obtain with the corresponding three-dimensional modeling data of property data base storehouse in three-dimensional model.The characteristic that offline feature is extracted each model in the computational data storehouse is passed through in the three-dimensional modeling data storehouse, and leaves these characteristics in property data base.Therefore, there is relation one to one in property data base and three-dimensional modeling data storehouse, and this feature extraction work is that off-line carries out, and does not take the time of on-line operation.Searching system generates the ordering of similarity according to the result of characteristic comparison, thereby according to the descending three-dimensional model that sorts of similarity, comes the most similar three-dimensional model of top representative, and sort method can adopt ripe methods such as quicksort.Searching system returns to the client user with the model tabulation of ordering through the network response; This tabulation is made up of two-dimentional thumbnail; Each thumbnail all corresponding a corresponding three-dimensional model, will return retrieval time of algorithm simultaneously, the user can browse to the two-dimentional thumbnail of similar three-dimensional model; After clicking two-dimentional thumbnail, can see the 3-D view of three-dimensional model, 3-D view all can slide through mouse and observe the detailed stereo content of three-dimensional model.
Be illustrated in figure 2 as the process of three-dimensional model feature extraction of the present invention.Earlier three-dimensional model is carried out dough sheet towards principal component analysis (PCA), three vertical main shafts of the dough sheet normal vector of computation model, obtain vertical with main shaft six accordingly towards sample plane.Evenly cut apart each towards sample plane according to some resolution; Launch towards the sampling ray towards sample plane from each; And generate towards sampling function, and this function is carried out multiresolution analysis obtains the multiresolution wavelet coefficient, final as towards multiresolution features.It should be noted that this characteristic is not single characteristic, but comprising the characteristic of a plurality of different resolutions.
In conjunction with accompanying drawing, elaborate below the practical implementation step.
One, adopt the dough sheet of 3D shape to obtain towards sample plane towards principal component analytical method.
The dough sheet that this step adopts 3D shape towards principal component analytical method obtain three towards main shaft, and six shape orientation sample plane are set according to these three vertical major, this sample plane spatially constitutes sealing, in 3D shape is enclosed in.This step need each dough sheet of Calculation of Three Dimensional shape towards the method for average vector of normal vector and 3D shape; The normal vector covariance matrix that constitutes thus; In this matrix; The normal vector of each dough sheet is carried out the weighting of area size, the eigenwert of rank order matrix and corresponding three proper vectors by size, thereby according to three proper vectors confirming as towards main shaft to calculate the sample plane of shape orientation.Be calculated as follows:
Suppose any 3D shape, total N dough sheet appointed and got a dough sheet T
i, the dough sheet area is f
i, represent dough sheet towards normal vector be n
i, this normal vector is through standardization, and the covariance matrix C that is made up of normal vector is:
Wherein f is the total surface area of 3D shape, and expression formula is:
N wherein
0Be all dough sheets towards method of average vector.
The covariance matrix of shape orientation is carried out characteristic operation, and formula is following:
Ce
i＝λ
ie
i
It is following to ask for eigenwert:
{λ
1，λ
2，λ
3}
Vectorial as follows with characteristic of correspondence:
{e
1，e
2，e
3}
And, arrange proper vector according to eigenwert order from big to small.
According to the sample plane of three proper vectors calculating shape orientations confirming, this sample plane satisfies two conditions:
(a) each proper vector is confirmed two sample plane, and this proper vector is vertical with two sample plane, and these two sample plane should lay respectively at the both sides of three-dimensional model, does not intersect with three-dimensional model, and three proper vectors are determined six sample plane altogether;
(b) in six sample plane are fully enclosed in three-dimensional model and since sample plane will gather 3D shape towards parameter, so sample plane apart the distance of three-dimensional model can set arbitrarily, this notion with bounding box is different.
Two, the sampling 3D shape towards distribution, generate towards the sampling function that distributes.
Sample plane through step (1) generates is evenly launched the sampling ray according to some levels; Intersect with the surface of 3D shape; The inner product of calculating sampling ray and the surface normal that intersects with it; With the functional value of inner product value as the shape orientation function, the index of employing 2 doubly carries out uniform sampling during sampling, is beneficial to adopt fast method that sampling function is analyzed.
Obtain towards the purpose that distributes and be: towards distribution is the key character of 3D shape dough sheet collection, the mankind can according to any 3D shape towards it is distinguished, towards representativeness with three-dimensional shape features.Have not with the far and near characteristics that change of sampled distance towards characteristic; This compares with the sampled distance function has robustness; If for example slight the variation takes place in a 3D shape; So three-dimensional bounding box will change, and the distance of the depth buffer of thereupon gathering will change, yet sampling is towards the influence that does not receive far and near distance.
Like Fig. 3, concrete grammar is following:
(1) each sample plane is divided into the little square of N * N, sampling resolution is N so, in order to carry out quick computing and easy analysis, N should for 2 arbitrary characteristics doubly, N=64 for example, and cut apart and want evenly;
(2) from ray V of center O emission of each little square, crossing with first dough sheet of 3D shape;
(3) obtain the normal vector L of first dough sheet, and with rays method vector V ask inner product of vectors (L ,-V), the reason that exists of negative sign is with the directions of rays negate here, can use the right-hand rule L to be rotated to-V.
(4) with inner product of vectors (L ,-V) value is given this little square center as sampled value, all little squares are all done the sampling computing.Two vector standardization in advance before assignment.
(5) can obtain like this a N * N two-dimentional uniform sampling function f (x, y).
The formula of final two-dimentional sampling function is following:
Because sample plane has six, therefore, sampling function also is made up of six corresponding functions, and is as follows:
f
o(x，y)，...，f
l(x，y)，...，f
s(x，y)，l?∈[0，5]
Wherein l representes l sample plane.
In actual operation, this invention has designed a kind of sampling accelerated method based on candidate's method, and Fig. 4 has provided this method.Among the figure; Appoint and get a triangle surface i, calculate 6 enveloping surfaces that square is formed in the sample plane, this enveloping surface surrounds this triangle minimum; Confirm 6 sampling rays that possibly intersect with it, two filled circles and four annulus are represented the starting point of 6 rays among Fig. 4.This method is no longer considered other sampling rays, and directly from the ray of 6 sampled point emissions, through 2 two real and triangle surface intersects rays of computational discrimination, the filled circles shown in the figure is the starting point of two rays.Calculate the sampling inner product value of corresponding 2 sampled points.This sampling accelerated method based on candidate's method can be saved the sampling time greatly.
Three, analytical sampling function, the multi-resolution low-frequency wavelet coefficient of extraction sampling function.
This step is selected the advantages of simplicity and high efficiency wavelet mother function; Owing to is function of region towards function, what relatively be good at for region-operation is generating function such as Daubechies, through the displacement and the change of scale of this generating function; Can constitute one group of function base; Utilize function base and according to scaling function and wavelet function to decomposing according to yardstick towards function, be that advanced every trade conversion, the method for carrying out rank transformation are again carried out dimensionality reduction for what two-dimentional sampling function adopted; Obtain the many groups low frequency wavelet coefficient under a plurality of resolution, the low frequency wavelet coefficient constitutes the characteristic of multiresolution.
The present invention in the characteristics of handling sampling function is, obtains the low resolution coefficient through the mode of wavelet decomposition, characterizes the characteristics of low-frequency that has towards sampling function.Obtaining characteristics of low-frequency has several different methods, for example Fourier analysis, and at first Fourier analysis can not provide multi-resolution mode, for the self-adaptation multiresolution features distance calculation of next step can not provide support; Also having a major reason in addition is that for two-dimensional function, the computation complexity of Fourier analysis is O (n
2Lgn), however the computation complexity of wavelet analysis is O (n
2), select for use Fourier analysis for very big influence was arranged on the whole three-dimensional model feature extraction time.Therefore the present invention adopts the wavelet decomposition mode to obtain the multi-resolution low-frequency characteristic.
Like Fig. 5, concrete steps are following:
(1) select the advantages of simplicity and high efficiency wavelet mother function, owing to be function of region towards function, what relatively be good at for region-operation is generating function such as Daubechies, through the displacement and the change of scale of this generating function, can constitute one group of function base;
(2) according to scaling function and wavelet function to decomposing according to yardstick towards function, generally be that advanced every trade conversion, the method for carrying out rank transformation are again carried out dimensionality reduction for what two-dimensional function adopted.
(3) obtain many groups low frequency wavelet coefficient under a plurality of resolution (for example 4-5).The dimension of every group of wavelet coefficient is 2
s* 2
s, s representes s resolution here.When s=1, converge 4 points, it is very few to comprise quantity of information, and feature representation is influenced to some extent, and the present invention begins to count from s=2, and the dimension of lowest resolution should be 4 * 4, if s=5, so high-resolution dimension should be 32 * 32, and the like.Can take out 5 groups of small echo low frequency coefficients under the resolution ratio in actual applications analyzes.
Because have six sample plane and six corresponding sampling functions, wavelet decomposition should be analyzed respectively each sampling function.
Four, through self-adaptation multiresolution features distance calculating method calculated characteristics distance.
This step is utilized the characteristic distance under certain resolution between two three-dimensional models of minimum Manhattan distance calculation of many sampling functions, utilizes the characteristic distance that the resolution method of weighting is calculated two three-dimensional models under all resolution simultaneously.The method of weighting that is adopted is adaptive weighted, utilizes sample database to make the First Tier parameter maximization in the search theory, to obtain optimum multiresolution features weighting coefficient.Make characteristic distance maximization and optimization through this apart from balancing method.
Designed a kind of self-adaptive features distance calculating method that is primarily aimed at multiresolution features in this invention; This method has different differentiation power according to the characteristic under the different resolution; Improve through adaptive weighted mode and to have certain resolution characteristics proportion of highly distinguishing power, thereby improve the distance regions calibration of whole multiresolution features.
Specifically be calculated as follows:
(1) characteristic distance under certain resolution between two three-dimensional models of minimum Manhattan distance calculation of the many sampling functions of employing.
Be located under any resolution k, the minimum Manhattan distance calculation that adopts many sampling functions between two three-dimensional models under this resolution apart from d
kWherein l representes l sample plane function, adopts six sample plane here altogether, can be surrounded as the cube of a sealing.In the specific implementation, also can adopt 12 or 24 sample plane to form the dodecahedron or the tetrahexahedrons of a sealing, purpose is can sample all shape orientations and do not omit.Distance between two three-dimensional models under the resolution k is:
d
k＝min{d
k，l}，l∈[0，5]
d
kGet minimum distance the distance between all sample plane functions.The distance of two three-dimensional model any sample plane function l under resolution k can be according to following characteristic distance computing formula:
d
kl＝||V
1，k，l-V
2，k，l||
V wherein
1, k, lThe small echo low frequency coefficient of representing first three-dimensional model l sample plane under resolution k, V
2, k, lThe small echo low frequency coefficient of representing second three-dimensional model l sample plane under resolution k.The small echo low frequency coefficient can be close with people's vision, therefore on similarity distance is measured, adopts the Manhattan distance.
(2) adopt the characteristic distance that the resolution method of weighting is calculated two three-dimensional models.
The separating capacity because the small echo low frequency coefficient under the different resolution takes on a different character; How to weigh the feature differentiation ability under a certain resolution; The present invention has designed two kinds of distance calculating methods, and first method is the fixed weighting method, and different characteristic under the different resolution is adopted fixed coefficient w
kWeighting; Second method is adaptive weighted method, calculates weighting coefficient w automatically according to query contents and classification
kThe defective of first method is can not draw optimum result to the retrieval of all types.Adopt second kind of adaptive weighted method to carry out adaptive weighted in the present invention to the wavelet coefficient characteristic under the different resolution.Method of weighting calculated distance formula is following:
(3) adopt adaptive weighted means, utilize sample database to make the First Tier parameter maximization in the search theory.
Provide First Tier computing method below.Suppose in the sample database D of three-dimensional model, to exist a three-dimensional model m
i, this three-dimensional model type of belonging to C
j, the model database size is that dimension is N, type C
jSize be that dimension is R, three-dimensional model m
iType of belonging to C
j, promptly
m
i∈C
j
This model of inquiry in this database, and will not comprise that the Query Result of this object itself sorts according to distance from small to large, promptly the most similar model comes the foremost.Only be concerned about now the model tabulation of ordering at preceding R-1, R is dimension of this ownership class, and it is not add up this object itself that R deducts 1 purpose.Supposing has P model to belong to same classification C in the model tabulation of preceding R-1
j, with respect to this model m
i,
FT
i＝P/(R-1)
FT wherein
iRepresent this model m
iFirst Tier parameter value.Desirable First Tier value should be 1.0, and when showing as the input searching object, preceding R-1 (not comprising input object itself) all belongs to same classification, explains that this characteristic distance has separating capacity the most completely.Under less than 1.0 situation, First Tier value is the bigger the better, and explains strong more apart from separating capacity.
Through the First Tier of all models of said method statistics, and obtain mean value to sample database dimension N.Mean F irst Tier does,
First Tier parameter in the formula is the result of calculation under some resolution.
Calculate the First Tier value under all resolution k then, compose and give w
kAs weight, thus distance weighted under the resolution k.Express as follows:
w
k＝FT
k
The final result calculated of this self-adaptation multiresolution features distance calculating method can be expressed as:
Adopt this adaptive weighted method to confirm that optimum characteristic distance distributes.When actual online retrieving, through calculating this characteristic distance between input model and the three-dimensional modeling data storehouse, similar three-dimensional data model can sort afterwards.
Claims (1)
1. based on the method for searching three-dimension model of shape orientation multi-resolution analysis, it is characterized in that comprising the steps:
(1) adopt the dough sheet of 3D shape to obtain towards sample plane towards principal component analytical method; May further comprise the steps: the dough sheet that adopts 3D shape towards principal component analytical method obtain three towards main shaft; And according to these three towards main shaft six shape orientation sample plane are set; This sample plane spatially constitutes sealing, in 3D shape is enclosed in; This step need each dough sheet of Calculation of Three Dimensional shape towards the method for average vector of normal vector and 3D shape; The normal vector covariance matrix that constitutes thus; In this matrix; The normal vector of each dough sheet is carried out the weighting of area size, the eigenwert of rank order matrix and corresponding three proper vectors by size, thereby according to three proper vectors confirming as towards main shaft to calculate the sample plane of shape orientation; Described sample plane satisfies two conditions: (a) each proper vector is confirmed two sample plane; This proper vector is vertical with two sample plane; And these two sample plane should lay respectively at the both sides of three-dimensional model; Do not intersect with three-dimensional model, three proper vectors are determined six sample plane altogether; (b) in six sample plane are fully enclosed in three-dimensional model and since sample plane will gather 3D shape towards parameter, so sample plane can be set arbitrarily apart from the distance of three-dimensional model;
(2) sampling 3D shape towards distribution; Generation may further comprise the steps towards the sampling function that distributes: evenly launch the sampling ray through sample plane, intersect with the surface of 3D shape; The inner product of calculating sampling ray and the surface normal that intersects with it; With the functional value of inner product value as the shape orientation function, the index of employing 2 doubly carries out uniform sampling during sampling, is beneficial to adopt fast method that sampling function is analyzed;
(3) analytical sampling function, the multi-resolution low-frequency wavelet coefficient of extraction sampling function may further comprise the steps:
Select the Daubechies generating function; Displacement and change of scale through this generating function; Constitute one group of function base; Utilize the function base and based on scaling function and wavelet function to decomposing according to yardstick towards function; The method that what adopt for two-dimentional sampling function is advanced every trade conversion, carry out rank transformation is again carried out dimensionality reduction; Obtain the many groups low frequency wavelet coefficient under a plurality of resolution ratio, the low frequency wavelet coefficient constitutes the characteristic of multiresolution;
(4) through self-adaptation multiresolution features distance calculating method calculated characteristics distance; May further comprise the steps: utilize the characteristic distance under certain resolution between two three-dimensional models of minimum Manhattan distance calculation of many sampling functions, utilize the characteristic distance that the resolution method of weighting is calculated following two three-dimensional models of all resolution simultaneously; The method of weighting that is adopted is adaptive weighted, utilizes sample database to make the First Tier parameter maximization in the search theory, to obtain optimum multiresolution features weighting coefficient.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2010102582422A CN101916284B (en) | 2010-08-19 | 2010-08-19 | Three-dimensional model searching method based on shape orientation multi-resolution analysis |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2010102582422A CN101916284B (en) | 2010-08-19 | 2010-08-19 | Three-dimensional model searching method based on shape orientation multi-resolution analysis |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN101916284A CN101916284A (en) | 2010-12-15 |
| CN101916284B true CN101916284B (en) | 2012-05-16 |
Family
ID=43323796
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN2010102582422A Expired - Fee Related CN101916284B (en) | 2010-08-19 | 2010-08-19 | Three-dimensional model searching method based on shape orientation multi-resolution analysis |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN101916284B (en) |
Families Citing this family (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102063475B (en) * | 2010-12-22 | 2012-10-10 | 张丛喆 | Webpage user terminal presenting method of three-dimensional model |
| CN105320764B (en) * | 2015-10-26 | 2019-05-07 | 天津大学 | A kind of 3D model retrieval method and its retrieval device based on the slow feature of increment |
| CN108549908B (en) * | 2018-04-13 | 2021-07-02 | 浙江科技学院 | Chemical process fault detection method based on multi-sampling probability kernel principal component model |
| CN110555219B (en) * | 2018-05-31 | 2022-12-27 | 首都航天机械公司 | Three-dimensional CAD model similarity retrieval system and method based on image recognition |
| CN110059213A (en) * | 2019-03-20 | 2019-07-26 | 杭州电子科技大学 | A kind of threedimensional model classification retrieving method based on Density Estimator |
| CN110737788B (en) * | 2019-10-16 | 2022-05-31 | 哈尔滨理工大学 | Rapid three-dimensional model index establishing and retrieving method |
| CN110930488B (en) * | 2019-11-15 | 2023-06-06 | 深圳市瑞立视多媒体科技有限公司 | Fish behavior simulation method, device, equipment and storage medium |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101196930A (en) * | 2008-01-04 | 2008-06-11 | 覃征 | Three-dimensional model searching system |
| CN101281545A (en) * | 2008-05-30 | 2008-10-08 | 清华大学 | Three-dimensional model search method based on multiple characteristic related feedback |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR100810326B1 (en) * | 2006-10-10 | 2008-03-04 | 삼성전자주식회사 | Method for generation of multi-resolution 3d model |
-
2010
- 2010-08-19 CN CN2010102582422A patent/CN101916284B/en not_active Expired - Fee Related
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101196930A (en) * | 2008-01-04 | 2008-06-11 | 覃征 | Three-dimensional model searching system |
| CN101281545A (en) * | 2008-05-30 | 2008-10-08 | 清华大学 | Three-dimensional model search method based on multiple characteristic related feedback |
Non-Patent Citations (2)
| Title |
|---|
| ZHENBAO LIU等.A New 3D Shape Retrieval Method Using Spherical Healpix.《Journal of Information Processing》.2008,第16卷190－200页. * |
| 盛洁谞,陈琪.多层多分辨率的扩展高斯图像.《武汉大学学报(理学版)》.2009,第55卷(第4期),486-490页. * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN101916284A (en) | 2010-12-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN101916284B (en) | Three-dimensional model searching method based on shape orientation multi-resolution analysis | |
| CN101271469B (en) | Two-dimension image recognition based on three-dimensional model warehouse and object reconstruction method | |
| Zhi et al. | LightNet: A Lightweight 3D Convolutional Neural Network for Real-Time 3D Object Recognition. | |
| Kazmi et al. | A survey of 2D and 3D shape descriptors | |
| CN108280187B (en) | Hierarchical image retrieval method based on depth features of convolutional neural network | |
| Zhang et al. | Pointwise geometric and semantic learning network on 3D point clouds | |
| CN109949368A (en) | A kind of human body three-dimensional Attitude estimation method based on image retrieval | |
| CN101201845B (en) | Method for searching three-dimensional model based on axis point-set hierarchical helix information | |
| CN106599053B (en) | Three-dimensional model retrieval method | |
| CN108875813A (en) | A kind of three-dimensional grid model search method based on several picture | |
| CN106844620B (en) | View-based feature matching three-dimensional model retrieval method | |
| CN105205135B (en) | A kind of 3D model retrieval methods and its retrieval device based on topic model | |
| CN110019914A (en) | A kind of three-dimensional modeling data storehouse search method for supporting three-dimensional scenic interaction | |
| CN101241504A (en) | Remote sense image data intelligent search method based on content | |
| CN105589974A (en) | Surveillance video retrieval method and system based on Hadoop platform | |
| CN113159232A (en) | Three-dimensional target classification and segmentation method | |
| CN111078916A (en) | Cross-domain three-dimensional model retrieval method based on multi-level feature alignment network | |
| CN105183795A (en) | Content based remote sensing image change detection information retrieval method | |
| CN106778639A (en) | A kind of exercise data search method based on the description of attitude relative space-time characteristic statisticses | |
| CN114743010A (en) | Ultrahigh voltage power transmission line point cloud data semantic segmentation method based on deep learning | |
| CN108596186B (en) | Three-dimensional model retrieval method | |
| CN111460193A (en) | Three-dimensional model classification method based on multi-mode information fusion | |
| CN113032613B (en) | Three-dimensional model retrieval method based on interactive attention convolution neural network | |
| CN114299339A (en) | Three-dimensional point cloud model classification method and system based on regional correlation modeling | |
| CN111597367B (en) | Three-dimensional model retrieval method based on view and hash 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 | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| ASS | Succession or assignment of patent right |
Owner name: NANTONG JINNIU MACHINERY MANUFACTURE CO., LTD. Free format text: FORMER OWNER: NORTHWESTERN POLYTECHNICAL UNIVERSITY Effective date: 20141014 Owner name: NORTHWESTERN POLYTECHNICAL UNIVERSITY Effective date: 20141014 |
|
| C41 | Transfer of patent application or patent right or utility model | ||
| COR | Change of bibliographic data |
Free format text: CORRECT: ADDRESS; FROM: 710072 XI'AN, SHAANXI PROVINCE TO: 226600 NANTONG, JIANGSU PROVINCE |
|
| TR01 | Transfer of patent right |
Effective date of registration: 20141014 Address after: 226600, Jiangsu province Nantong city Haian county old dam Town Industrial Park Patentee after: NANTONG JINNIU MACHINERY MANUFACTURE CO., LTD. Patentee after: Northwestern Polytechnical University Address before: 710072 Xi'an friendship West Road, Shaanxi, No. 127 Patentee before: Northwestern Polytechnical University |
|
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20120516 Termination date: 20190819 |