A kind of compressed sensing based method for extracting characteristics of three-dimensional model
Technical field
The present invention relates to threedimensional model process fields, more specifically to a kind of compressed sensing based threedimensional model
Feature extracting method.
Background technique
With the rapid development of information retrieval technique and the raising of computer performance, information processing is from traditional mode to new
Type Mode change.Relative to text information and two dimensional image, more true threedimensional model abundant using more and more extensive.?
Nowadays in magnanimity 3 d model library, how to realize the management and retrieval reused based on threedimensional model, rapidly and accurately find and meet
It is required that threedimensional model, it has also become an important research topic of searching field at present.
As the 4th kind of multimedia data type after sound, image and video, the threedimensional model inspection based on content
The development of rope technology is concerned.How the feature but also efficiently and accurately measurement model similitude of content had not only quickly and easily been extracted
It is the key problem in the three-dimensional model search technology based on content, both of these problems are the hot issues being widely studied,
It is extremely challenging one of difficulties.
Method for searching three-dimension model of the majority based on content there is also some problems at present: as extraction feature cannot be complete
Expression three-dimensional model information, computation complexity are high, feature extraction and the time of characteristic matching are long, characteristic storage space is big, feature
Information is easy to lack, can not achieve user interactive operation etc..And in method for searching three-dimension model, feature is carried out to threedimensional model
Extraction is to guarantee the important means of retrieval rate and quality, therefore, as multimedia application field is to three-dimensional model search speed
How the requirement being continuously improved with quality improves the Accuracy and high efficiency for carrying out feature extraction to threedimensional model, is three-dimensional mould
Type searching field needs deeper into research and the project explored.
Summary of the invention
It is an object of the invention to overcome shortcoming and deficiency in the prior art, a kind of compressed sensing based three-dimensional is provided
Aspect of model extracting method, the method for extracting characteristics of three-dimensional model can multi-angle reflection threedimensional model feature, realize to voxel
The threedimensional model for changing format carries out space delamination processing, carries out spatial decomposition to the threedimensional model of labyrinth, not only improves three
The Accuracy and high efficiency of dimension module feature extraction, and the efficient space geometry feature of low-dimensional is extracted, avoid feature redundancy.
In order to achieve the above object, the technical scheme is that: one kind compressed sensing based three
Dimension module feature extracting method, it is characterised in that:
Firstly, choose the threedimensional model of discrete voxelization format, then choose the orientation at each visual angle as a reference plane, and
Contour transforming function transformation function is designed, threedimensional model is realized into space delamination by contour transforming function transformation function, obtains space delamination model;
Secondly, each space delamination model projection to reference planes is constructed projection matrix, and extract the letter of projection matrix
Cease entropy;
Finally, carrying out sparse processing to each projection matrix, and two dimensional compaction perception processing is carried out, obtains space delamination spy
Sign.
In the above scheme, the present invention can multi-angle reflection by compressed sensing based method for extracting characteristics of three-dimensional model
The feature of threedimensional model is realized and carries out space delamination processing to the threedimensional model of voxelization format, to the three-dimensional mould of labyrinth
Type carries out spatial decomposition, to improve the Accuracy and high efficiency of threedimensional model feature extraction, and then guarantees three-dimensional model search
Speed and quality.
The method of the present invention the following steps are included:
Step s101: it chooses threedimensional model and is the threedimensional model of discrete voxelization format, and carry out threedimensional model voxelization
Pretreatment, obtains the pretreated threedimensional model M of voxelization (s × s × s), and wherein s is discrete voxel model resolution ratio;
Step s102: using xoy=0 plane as reference plane, select contour transforming function transformation function mapping: f (x, y, z)=z is drawn
Dividing hierarchy number is L, and layering step-length is step=s/L;Construct L projection matrix projl(BS × BS), l=1,2 ... L;BS is
The size of projection matrix, each element of matrix is 0 when initial, i.e. projl(i, j)=0, i.j=1,2 ..., BS;L=1,
2,...,L;
Step s103: to tissue points any in modelIt is calculated by following formula and updates throwing
The element of shadow matrix:
Wherein l is k-th of tissue points vkThe affiliated number of plies, the i.e. element of projection matrix;
Step s104: firstly, calculating l layers of projection matrix projlComentropy componentIts
It is secondary, obtain the comentropy of Z-direction
Step s105: to projection matrix projlUsing DCT sparse transformation, sparse signal χ is obtainedl:
χl=DCT (projl);
Step s106: to sparse signal χlIt is perceived using two dimensional compaction, obtains compressed sensing measuring signal γl:
Wherein Φ1, Φ2For calculation matrix;
Step s107: calculating 2 norms of the measured value of each layering, forms characteristic sequence FZ:
Step s108: similarly similar s102~s107 is for reference plane and is operated with yoz=0, xoz=0 respectively, obtains spy
Levy sequence FX, FY;
Step s109: merging the characteristic sequence in three directions, exports the spatial scalable compression based on functional transformation and perceives spy
Levy FHCS(Hierarchical CS):
Step s110: merging the Information Entropy Features sequence in three directions, exports space delamination Information Entropy Features FENT:
Specifically, the selection reference planes are the corresponding plane yoz of space coordinates, plane xoz and plane xoy.
The space delamination feature includes compressed sensing feature FHCSWith layered entropy feature FENT。
In implementation process, selected model is the threedimensional model of discrete voxelization format, which is a kind of entity
Model can reflect model internal information.In addition it is directed to threedimensional model characteristic in the program, chooses the basic engineering etc. of different direction
High transforming function transformation function, multi-angle reflect the feature of threedimensional model, realize and handle the space delamination of voxel model, to labyrinth
Threedimensional model carries out spatial decomposition.Then each space layer of threedimensional model is projected according to reference ground, is projected
Matrix, the projection matrix have the depth characteristic of hierarchical mode, i.e. the information of projection matrix is extracted in the performance of the hypostazation of model
Entropy obtains the situation of change of Information Entropy Features sequence.Sparse processing is carried out to each projection matrix, avoids the not sparse institute's band of signal
Carry out the difficult problem of compressed sensing reconstruct.Two dimensional compaction perception finally is carried out to each sparse signal, extracts compressed sensing feature,
The characteristic sequence of low dimensional is formed, the characteristic sequence of different direction can effectively complete expression model.
Compared with prior art, the invention has the advantages that with the utility model has the advantages that the present invention is based on the three-dimensionals of compressed sensing
Aspect of model extracting method can multi-angle reflection threedimensional model feature, realize and space carried out to the threedimensional model of voxelization format
Layered shaping carries out spatial decomposition to the threedimensional model of labyrinth, not only improve threedimensional model feature extraction accuracy and
High efficiency, and the efficient space geometry feature of low-dimensional is extracted, feature redundancy is avoided, to guarantee the speed of three-dimensional model search
And quality.
Detailed description of the invention
Fig. 1 is the flow chart of the method for extracting characteristics of three-dimensional model the present invention is based on compressed sensing;
Specific embodiment
The present invention is described in further detail with specific embodiment with reference to the accompanying drawing.
Embodiment
As shown in Figure 1, the method for extracting characteristics of three-dimensional model the present invention is based on compressed sensing is such that
Firstly, choose the threedimensional model of discrete voxelization format, then choose the orientation at each visual angle as a reference plane, and
Contour transforming function transformation function fi is designed, threedimensional model is realized into space delamination by contour transforming function transformation function fi, obtains space delamination model;
Secondly, each space delamination model projection to reference planes is constructed projection matrix, and extract the letter of projection matrix
Cease entropy;
Finally, carrying out sparse processing to each projection matrix, and two dimensional compaction perception processing is carried out, obtains space delamination spy
Sign.
This method specifically includes the following steps:
Step s101: it chooses threedimensional model and is the threedimensional model of discrete voxelization format, and carry out threedimensional model voxelization
Pretreatment, obtains the pretreated threedimensional model M of voxelization (s × s × s), and wherein s is discrete voxel model resolution ratio;
Step s102: using xoy=0 plane as reference plane, selecting contour transforming function transformation function fi mapping: f (x, y, z)=z,
Division hierarchy number is L, and layering step-length is step=s/L;Construct L projection matrix projl(BS × BS), l=1,2 ... L;BS
For the size of projection matrix, each element of matrix is 0 when initial, i.e. projl(i, j)=0, i.j=1,2 ..., BS;L=
1,2,...,L;
Step s103: to tissue points any in modelIt is calculated by following formula and updates throwing
The element of shadow matrix:
Wherein l is k-th of tissue points vkThe affiliated number of plies, the i.e. element of projection matrix;
Step s104: firstly, calculating l layers of projection matrix projlComentropy componentIts
It is secondary, obtain the comentropy of Z-direction
Step s105: to projection matrix projlUsing DCT sparse transformation, sparse signal χ is obtainedl:
χl=DCT (projl);
Step s106: to sparse signal χlIt is perceived using two dimensional compaction, obtains compressed sensing measuring signal γl:
Wherein Φ1, Φ2For calculation matrix;
Step s107: calculating 2 norms of the measured value of each layering, forms characteristic sequence FZ:
Step s108: similarly similar s102~s107 is for reference plane and is operated with yoz=0, xoz=0 respectively, obtains spy
Levy sequence FX, FY;
Step s109: merging the characteristic sequence in three directions, exports the spatial scalable compression based on functional transformation and perceives spy
Levy FHCS(Hierarchical CS):
Step s110: merging the Information Entropy Features sequence in three directions, exports space delamination Information Entropy Features FENT:
Wherein, the present invention chooses reference planes as the corresponding plane yoz of space coordinates, plane xoz and plane xoy, and
Space delamination feature includes compressed sensing feature FHCSWith layered entropy feature FENT。
In implementation process, selected model is the threedimensional model of discrete voxelization format, which is a kind of entity
Model can reflect model internal information.In addition it is directed to threedimensional model characteristic in the program, chooses the basic engineering etc. of different direction
High transforming function transformation function fi, multi-angle reflect the feature of threedimensional model, realize and handle the space delamination of voxel model, to labyrinth
Threedimensional model carry out spatial decomposition.Then each space layer of threedimensional model is projected according to reference ground, is thrown
Shadow matrix, the projection matrix have the depth characteristic of hierarchical mode, i.e. the letter of projection matrix is extracted in the performance of the hypostazation of model
Entropy is ceased, the situation of change of Information Entropy Features sequence is obtained.Sparse processing is carried out to each projection matrix, avoids the not sparse institute of signal
The problem for bringing compressed sensing reconstruct difficult.Two dimensional compaction perception finally is carried out to each sparse signal, it is special to extract compressed sensing
Sign, forms the characteristic sequence of low dimensional, the characteristic sequence of different direction can effectively complete expression model.
The present invention is based on the method for extracting characteristics of three-dimensional model of compressed sensing can multi-angle reflection threedimensional model feature, it is real
Space delamination processing now is carried out to the threedimensional model of voxelization format, spatial decomposition is carried out to the threedimensional model of labyrinth, no
The Accuracy and high efficiency of threedimensional model feature extraction is only improved, and extracts the efficient space geometry feature of low-dimensional, avoids spy
Redundancy is levied, to guarantee the speed and quality of three-dimensional model search.
The above embodiment is a preferred embodiment of the present invention, but embodiments of the present invention are not by above-described embodiment
Limitation, other any changes, modifications, substitutions, combinations, simplifications made without departing from the spirit and principles of the present invention,
It should be equivalent substitute mode, be included within the scope of the present invention.