CN104252715B - Single line image-based three-dimensional reconstruction method - Google Patents
Single line image-based three-dimensional reconstruction method Download PDFInfo
- Publication number
- CN104252715B CN104252715B CN201410450000.1A CN201410450000A CN104252715B CN 104252715 B CN104252715 B CN 104252715B CN 201410450000 A CN201410450000 A CN 201410450000A CN 104252715 B CN104252715 B CN 104252715B
- Authority
- CN
- China
- Prior art keywords
- lines
- dimensional reconstruction
- line
- coordinate
- model
- 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
Landscapes
- Image Analysis (AREA)
Abstract
The invention discloses a single line image-based three-dimensional reconstruction method. The method comprises the following steps: firstly performing vectorization processing on an input line image, and converting the line image into a two-dimensional vector line plot; then, matching the two-dimensional vector line plot and a preset three-dimensional model base by applying a subgraph isomorphism method, wherein the model matched with the two-dimensional vector line plot is called a candidate model; finally, performing minimized solving on the coordinate difference function of the line plot and the candidate model so as to select the optimal model and obtain a reconstruction result. According to the single line image-based three-dimensional reconstruction method disckosed by the invention, three-dimensional reconstruction can be effectively performed on the single line image, so that the reading experiences of related documents are greatly improved.
Description
Technical field
The invention belongs to image procossing and computer vision field, are related to a kind of 3 D stereo based on single width lines image
Method for reconstructing.
Background technology
Substantial amounts of three-dimensional geometrical figure is there is in electronic document miscellaneous.These documents include teaching material, examination question,
Demonstration original text etc..However, these three-dimensional geometrical figure overwhelming majority are stored in a document in the way of two-dimentional lines image, and nothing
Method is directly presented the 3-D solid structure of object, and the reading to reader causes inconvenience.Particularly in mobile reading equipment, three-dimensional
Display device more and more ripe today, the three-dimensional geometrical figure for being stored with two-dimentional lines image mode and being shown have already fallen behind in
Advanced Display Technique, have impact on the reading experience of reader.If can be 3 D stereo by these lines image restoration and reconstruction,
The reading experience of reader will be greatly enhanced.
" lines figure " is a kind of artificial X-Y scheme, and it can describe the structure of object in the way of clear and intuitive.No
General natural image is same as, the structure of lines figure is often made up of point, line, with clearly geometry, by parallel projection
Constitute, no see-through feature, not with the texture information in general nature image.The common lines figure of people includes manual draw
Shape, Graphing of Engineering, CAD drawing, teaching material illustration, etc..Actually one connected graph containing summit and side of lines figure (for
Unconnected graph can be divided into connected subgraph and be respectively processed).One is generally required to the algorithm that lines figure carries out three-dimensional reconstruction
Individual pretreatment is converted to summit two-dimensional coordinate and connection line set figure.Connection side between summit, can be straightway,
It can be curve.
In past 20 years, occur in that the research of a large amount of three-dimensional reconstruction methods based on single width lines figure.These researchs are adopted
Method, applicable object, application scenarios are all not quite similar.By method for reconstructing adopt algorithm classification, have based on Optimality Criteria,
Based on geometry hypothetical deduction, based on methods such as divide-and-conquer strategies.Rule-based early stage, most of method was adopted, and they are by three
Dimension solid is defined as geometric object, and which is used restraint with some geometrical rules, to seek optimal knot in solution room
Really.Conventional rule has plane rule (document " An optimization-based approach to the
interpretation of single line drawings as 3D wire frames.International
Journal of Computer Vision, 1992. "), MSDA rule (document " Emulating the human
interpretation of line-drawings as three-dimensional objects.International
Journal of Computer Vision, 1991. "), MSDSM rule (document " 3D Object Recovery from 2D
Images：A New Approach.SPIE Proc.Robotics and Computer Vision, 1996. "), MEAD is regular
(document " 3-D interpretation of single line drawings based on entropy
Minimization principle.Computer Vision and Pattern Recognition, 2001. ") etc..Solve
Majorized function is usually the final step of this kind of method；The method assumed based on geometry assumes initially that the 3 D stereo handled by which
Certain requirement is followed, such as contains cube angle point (document " 3D reconstruction of polyhedral objects
From single parallel projections using cubic corner.Computer-Aided Design,
2011. "), there are spatial symmetry (document " Inferring mirror symmetric 3D shapes from
Sketches.Computer-Aided Design, 2012. ") etc..On the basis of hypothesis, their points to whole object and
Side coordinate speculated, finally gives stereo reconstruction result.Optimization Solution step is included in this method not necessarily, it is general to calculate
Complexity is relatively low, but its assumed condition is generally harsher, is only applicable to some special figures；In recent years some sides for occurring
Method (document " A divide-and-conquer approach to 3D object reconstruction from line
Drawings.IEEE 11th International Conference on Computer Vision, 2007. ", document
“Decomposition of complex line drawings with hidden lines for 3D planar-faced
manifold object reconstruction.IEEE Transactions on Pattern Analysis and
Machine Intelligence, 2011. ", document " Object cut：Complex 3d object reconstruction
through line drawing separation.IEEE Conference on Computer Vision and
Pattern Recognition, 2010. ") employ the strategy divided and ruled to solve the problems, such as some complicated three-dimensional reconstructions,
Complicated figure is first divided into simple essential part as far as possible by them, then using some rules being optimized solution, most
Afterwards the result that various pieces are tried to achieve is combined and obtains the result of whole reconstruction.
It can be seen that, the current three-dimensional reconstruction method to single width lines figure can only process the lines figure of vector quantization mostly,
And image cannot be processed.Even if lines image can be converted into the lines figure of vector, mesh by certain pretreatment
Front method also highly relies on the integrity and correctness of lines figure, and some even have to rely on manual process and extract lines figure, and
Do not have it is a kind of it is efficient, can directly process image, the three-dimensional reconstruction method that has fault-tolerant ability.
The content of the invention
In order to realize that efficient three-dimensional reconstruction is carried out to single width lines image, the present invention proposes a kind of based on single width line
The three-dimensional reconstruction method of bar image.The lines image of input is carried out vectorized process by the method first, is converted into
Two-dimensional vector lines figure, then using the method for " subgraph match " by two-dimensional vector lines figure and 3 d model library set in advance
Matched, matching model is referred to as candidate family, finally the coordinate difference by lines figure with candidate family is carried out away from function
Minimize and solve, to select the model of optimum and draw reconstructed results.The present invention effectively can be carried out to single width lines image
Three-dimensional reconstruction.
The process object of the present invention is the geometry lines image extracted from the electronic document of PDF or other forms, or logical
Cross mobile phone and other camera installations shoot, or the geometry lines image obtained by scanner scanning papery teaching material.
The technical scheme that the present invention is provided is as follows：
A kind of three-dimensional reconstruction method based on single width lines image, is characterized in that, comprise the steps：
1) vector lines figure is extracted from input picture；
2) some candidate families are chosen for vector lines figure from 3 d model library；
3) apex coordinate of vector lines figure is fixed, in three dimensions rotation, Pan and Zoom candidate family with
2 D vertex coordinate with vector lines figure so that vector lines figure is reached most with the variance of the coordinate on each summit of candidate family
Little value；Then the minimum model of the variance of coordinate is selected from several candidate families, as the result of three-dimensional reconstruction.
Preferably：
Described three-dimensional reconstruction method, is characterized in that, step 1) implementation method be：
1.1) input picture is carried out into binary conversion treatment and connected component's search；
1.2) lines to be extracted in image are divided into solid line and dotted line carries out the extraction of straight line；
1.3) line segment is cut into straight line according to the intersection point of the straight line for being extracted；
1.4) unnecessary lines are filtered out, vector lines figure is obtained.
Described three-dimensional reconstruction method, is characterized in that, step 2) described in 3 d model library in, with parametrization shape
Formula preserves threedimensional model：One threedimensional model is the lines figure in a three dimensions, and its apex coordinate is by one group of state modulator.
Described three-dimensional reconstruction method, is characterized in that, step 2) in, the use of the method for Subgraph Isomorphism is vector lines
Figure chooses some candidate families.
Described three-dimensional reconstruction method, is characterized in that, step 2) in, need to be matched twice, for the first time by vector
Lines figure is schemed as big, using threedimensional model as little figure；Second using threedimensional model as big figure, using vector lines figure as little
Figure.
Described three-dimensional reconstruction method, is characterized in that, step 1) implementation method be：
1.1) k-means clusters are carried out to the connected component of input picture, using the encirclement frame size and picture of connected component
Vegetarian noodles product clusters attribute as which, is divided three classes：Main frame, dotted line point, descriptive text；
1.2) solid line extraction is carried out to main frame connected component using Hough transform, dotted line is carried out using RANSAC methods
Extract；
1.3) remove hanging line, stop the unnecessary lines such as line, diagonal, generate vector lines figure.
Described three-dimensional reconstruction method, is characterized in that, step 3) implementation method be：
3.1) vector lines figure apex coordinate is fixed, in three dimensions rotation, Pan and Zoom candidate family with
2 D vertex coordinate with vector lines figure；
3.2) coordinate of the candidate family through parallel projection to vector lines plan is calculated, and obtains model projection coordinate
With the variance of lines figure apex coordinate, constitute object function and solution is optimized to which；
3.3) candidate family for causing object function minimum is chosen as optimal models.
Described three-dimensional reconstruction method, is characterized in that, further comprising the steps of：
4) export the result of three-dimensional reconstruction.
Effect of the invention is that：Realize a kind of three-dimensional reconstruction method based on single width lines image.By right
Input picture extracts two-dimensional vector lines figure, and which is matched with 3 d model library set in advance, selects out candidate family,
Obtain the result of three-dimensional reconstruction again by the rotation to candidate family, Pan and Zoom.The method can solve the problem that current side
Method directly can not process image, to be input into lines figure integrity demands it is high the shortcomings of, treatment effeciency can be lifted and abundant moved
The reading experience of equipment user.
Description of the drawings
Fig. 1 is the flow process frame diagram of the present invention；
Fig. 2 is the method flow diagram that vector lines figure is extracted from image of the present invention；
Fig. 3 is the schematic diagram that vector lines figure is extracted from image.A () input picture (b) is carried to connected component's cluster (c)
Treating excess syndrome line (d) extracts dotted line (e) resultant vector lines figure；
Fig. 4 is three terrace with edge of default 3 d model library example (a) cuboid (b) rectangular pyramid (c) triangular prism (d)；
Fig. 5 is that the method schematic diagram (a) dotted line point (b) (c) for extracting dotted line extracts straight line (d) outlier；
Fig. 6 is the schematic diagram that unnecessary lines are filtered；
Input file and pictures of the Fig. 7 for specific embodiment；
Result schematic diagrams of the Fig. 8 for vector extraction lines figure；
Schematic diagrams of the Fig. 9 for Subgraph Isomorphism matching candidate model；
Figure 10 is three-dimensional reconstruction output result schematic diagram；
Specific embodiment
The tool of the present invention is introduced below for using file and picture as the application scenarios of the input of three-dimensional reconstruction system
Body implementing procedure.Here file and picture is the geometry lines image extracted from the electronic document of PDF or other forms, or logical
Cross mobile phone and other camera installations shoot, or the geometry lines image obtained by scanner scanning papery teaching material.User U's sets
Standby (PC, handheld device etc.) needs first to install the execution software for realizing function of the present invention, and after installing, user U can
To open PDF document in software, by mouse or gesture positioning and intercepting image, 3 D stereo weight is performed automatically in software
Build work output result.User U can be watched on screen and be rotated the 3-D graphic of output.
The specific implementation step of the present invention is (ginseng Fig. 1)：
(1) extract vector lines figure
The flow process of method is as shown in Figure 2.For the solid line in image, the maximum component in image is carried out into Hough
Conversion process；For the dotted line in image, less connected component in image is screened by K-means clusters, then
Extraction straight line is carried out by RANSAC methods.Then the intersection point of extracted straight line is calculated, is line segment straight line cutting from point of intersection,
And merge adjoining nodes.Eventually pass filtration treatment and remove unnecessary lines, generate final line segment and vector lines figure, Fig. 3 according to
It is secondary to show file and picture example, solid line, the extraction result of dotted line, and the vector lines figure for generating.It is below concrete implementation
Method：
1.1) connected component's cluster.As shown in Fig. 3 (b), in file and picture, generally there is the connected component of three types：Main frame
Frame, dotted line point, and descriptive text.The connected component in image is clustered using k-means methods.If k=3, and
The encirclement frame size and elemental area of connected component is taken to cluster attribute as which.After a wheel cluster, successfully will even
The reduction of fractions to a common denominator is divided into three classes, and the wherein maximum class of area is chosen as main frame, and the similar and minimum class of area is dotted line point,
Other is descriptive text.
1.2) solid line is extracted.Hough transform and the method based on edge line segment are employed on the Canny edges of main frame
Carry out lines detection.As a line on image has two Canny edges, therefore the straight line that a line is extracted has two simultaneously
And with almost identical slope and very near distance.These straight lines are merged into into one according to their slope and distance relation
Bar.
1.3) dotted line is extracted.As shown in figure 5, carrying out the extraction of dotted line using RANSAC methods.First, it is all to be divided
Connected component for dotted line point is reduced into its central point, then can determine that straight line per 2 points.Find in these straight lines
Comprising the most straight line of interior point (a sufficiently small scope is less than with a distance from straight line), take out this straight line and it is included
Interior point reject.Repeat above procedure until can not find the straight line that the above is put in comprising 3.Remaining point becomes outlier.
1.4) unnecessary lines are filtered.For some unnecessary lines (boost line, lines that mistake is extracted etc.), need
They are removed to improve the success rate of Model Matching.The concrete line style of type for removing is as follows：
A) hanging line：In the vector lines figure for extracting, if the degree of a lines end points is 1, it is referred to as hanging line.
If the lines 6-7 in Fig. 6 (a) is exactly a hanging line, the degree of its end points 7 is 1.Other typical hanging lines are more common in coordinate
The vector lines figure of axle and imperfect extraction.For hanging line needs to remove them.
B) stop line：If just the mid portion of another lines (not being two ends), it is one to the end points of a lines
Bar stops line (Fig. 6 (b) CE, CF, C1E, C1F).Boost line in many file and pictures is all to stop line.These lines are also required to
It is removed.
C) diagonal：Another kind of unnecessary lines are the diagonal of the parallelogram in figure, A in such as Fig. 6 (c)1B, BC1,
A1C1.What these diagonal destroyed script object opens up benefit structure, it is therefore desirable to be removed.
(2) threedimensional model matching
Based on the analysis to representative document image, the present invention establishes following 3 d model library：If 3 d model library by
Dry threedimensional model composition, a model is the lines figure in a three dimensions, and its apex coordinate is by one group of state modulator
's.Model in 3 d model library is all the typical solid figure in file and picture, such as cuboid, rectangular pyramid, three terrace with edges etc., such as
Shown in Fig. 4.By taking Fig. 4 (a) as an example, a rectangular body Model has three parameters：A={ x, y, z }, represents this with parameter matrix V
All apex coordinates of individual model are
Each row of wherein V represent the three-dimensional coordinate on a summit, and a dimension, controls of a for model are represented per a line
Parameter vector, it controls the geometric attributes such as the length of model.It is any comprising NmThe model m on individual summit has 3 × NmGinseng
Matrix number Vm。
After the completion of vector lines figure is extracted, one group of candidate's mould is found out in model library using VF-2 Subgraph Isomorphism algorithms
Type." Subgraph Isomorphism " is a kind of technology for opening up the consistent subgraph of benefit structure found in a big figure with a little figure.
The present invention is required for carrying out Matching Model twice every time：" positive matching " and " inversely matching ".What positive matching referred to
It is, as big figure, using threedimensional model as little figure, the mode of the subgraph of isomorphism to be found in vector lines figure using vector lines figure.
And then contrast is inversely matched, it is that threedimensional model is schemed as big, and using vector lines figure as little figure, in threedimensional model
Middle searching isomorphism subgraph.
Positive matching way is only used for the vector lines figure of complete extraction.Due to the office of vector lines figure extraction algorithm
Limit, for incomplete vector lines figure (such as Fig. 6 (a)) for extracting, positive matching will fail.Therefore, using reverse match party
Method is solving this problem：Threedimensional model finds vector lines graph isomorhpism subgraph wherein as big figure.Therefore, reverse
In matching, even if vector lines figure is imperfect, candidate family also can be successfully found out.
(3) three-dimensional reconstruction
Based on the result that threedimensional model is matched, adopt：By vector lines figure summit
Coordinate is fixed, and rotation, Pan and Zoom model are matching the 2 D vertex coordinate of vector lines figure in three dimensions.Concrete side
Method is：One three-dimensional spin matrix R of definition, translation matrix t, the control parameter of model is a, if institute's Matching Model
Each apex coordinate be Vi, (scaling of apex coordinate is by a controls), the apex coordinate of vector lines figure is xi, then threedimensional model
Through rotation, translation after coordinate be
(RVi+t) (1)
So it with the coordinate difference on each summit of former vector lines figure is
(RVi+t)-xi (2)
Target be find it is suchSo that the quadratic sum of the coordinate difference on all summits is minimum, i.e.,
Using document " A feasible method for optimization with orthogonality
Constraints.Technical report, Rice University, 2010 " method solves relevant optimization problem.Before
Each candidate family for selecting can carry out the parameter optimization of formula (3), obtain optimum resultsWherein so that target
Functional value f minimum model is finalized as whole modeling type.Final reconstructed results are obtained according to optimization by whole modeling type
ParameterCarry out the rotation in three dimensions, Pan and Zoom to obtain
Three-dimensional coordinates of the wherein S for each summit of output result,For spin matrix, V be byElement constitute parameter
Matrix (The scaling of control vertex coordinate), T serves as reasonsThe translation matrix of composition.
Specific embodiment
Below according to above-mentioned specific implementation method, for the process that the file and picture of width input, the description present invention are realized.
Intercept a width geometry lines image first from the electronic document of PDF or other forms, or pass through mobile phone and other
Camera installation shoots, or obtains a width geometry lines image to the scanning of papery teaching material by scanner, the image that then will be obtained
The input of (as shown in Figure 7) as system.The result for extracting vector lines figure is as shown in Figure 8.Then selected using Subgraph Isomorphism
The candidate family (as shown in Figure 9) of matching, and only retain the matching result of clique.Eventually through rotation, translation candidate's mould
Type, and choose the minimum result of optimization object function and complete three-dimensional reconstruction.Final output result is as shown in Figure 10.
Claims (6)
1. a kind of three-dimensional reconstruction method based on single width lines image, is characterized in that, comprise the steps：
1) vector lines figure is extracted from input picture；
2) method of the Subgraph Isomorphism from used in 3 d model library is that vector lines figure chooses some candidate families；And two need to be carried out
Secondary matching, for the first time using vector lines figure as big figure, using threedimensional model as little figure；Second using threedimensional model as big
Figure, using vector lines figure as little figure；
3) apex coordinate of vector lines figure is fixed, rotation, Pan and Zoom candidate family are matching arrow in three dimensions
The 2 D vertex coordinate of amount lines figure so that vector lines figure reaches minimum with the variance of the coordinate on each summit of candidate family
Value；Then the minimum model of the variance of coordinate is selected from several candidate families, as the result of three-dimensional reconstruction.
2. three-dimensional reconstruction method as claimed in claim 1, is characterized in that, step 1) implementation method be：
1.1) input picture is carried out into binary conversion treatment and connected component's search；
1.2) lines to be extracted in image are divided into solid line and dotted line carries out the extraction of straight line；
1.3) line segment is cut into straight line according to the intersection point of the straight line for being extracted；
1.4) unnecessary lines are filtered out, vector lines figure is obtained.
3. three-dimensional reconstruction method as claimed in claim 1, is characterized in that, step 2) described in 3 d model library in,
Threedimensional model is preserved with parameterized form：One threedimensional model is the lines figure in a three dimensions, and its apex coordinate is by one
Group state modulator.
4. three-dimensional reconstruction method as claimed in claim 1, is characterized in that, step 1) implementation method be：
1.1) k-means clusters are carried out to the connected component of input picture, using the encirclement frame size and pixel faces of connected component
Product clusters attribute as which, is divided three classes：Main frame, dotted line point, descriptive text；
1.2) solid line extraction is carried out to main frame connected component using Hough transform, dotted line extraction is carried out using RANSAC methods；
1.3) remove including hanging line, the unnecessary lines stopped including line, diagonal, generate vector lines figure.
5. three-dimensional reconstruction method as claimed in claim 1, is characterized in that, step 3) implementation method be：
3.1) vector lines figure apex coordinate is fixed, rotation, Pan and Zoom candidate family are matching arrow in three dimensions
The 2 D vertex coordinate of amount lines figure；
3.2) coordinate of the candidate family through parallel projection to vector lines plan is calculated, and obtains model projection coordinate and line
The variance of bar figure apex coordinate, constitutes object function and solution is optimized to which；
3.3) candidate family for causing object function minimum is chosen as optimal models.
6. three-dimensional reconstruction method as claimed in claim 1, is characterized in that, further comprising the steps of：
4) export the result of three-dimensional reconstruction.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410450000.1A CN104252715B (en) | 2014-09-05 | 2014-09-05 | Single line image-based three-dimensional reconstruction method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410450000.1A CN104252715B (en) | 2014-09-05 | 2014-09-05 | Single line image-based three-dimensional reconstruction method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104252715A CN104252715A (en) | 2014-12-31 |
| CN104252715B true CN104252715B (en) | 2017-05-03 |
Family
ID=52187581
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410450000.1A Active CN104252715B (en) | 2014-09-05 | 2014-09-05 | Single line image-based three-dimensional reconstruction method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104252715B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9646411B2 (en) * | 2015-04-02 | 2017-05-09 | Hedronx Inc. | Virtual three-dimensional model generation based on virtual hexahedron models |
| CN106004140B (en) * | 2016-05-19 | 2018-01-09 | 清华大学 | The method that 3D 3-D cartoons are shown in single image |
| CN107871333B (en) * | 2017-11-30 | 2021-04-16 | 上海联影医疗科技股份有限公司 | Line drawing method and device, computer equipment and computer storage medium |
| CN110414402B (en) * | 2019-07-22 | 2022-03-25 | 北京达佳互联信息技术有限公司 | Gesture data labeling method and device, electronic equipment and storage medium |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101937579A (en) * | 2010-09-20 | 2011-01-05 | 南京大学 | Method for creating three-dimensional surface model by using perspective sketch |
| CN102880868A (en) * | 2012-08-06 | 2013-01-16 | 上海中和软件有限公司 | Engineering drawing vector conversion and primitive semantic extraction method |
| CN103390088A (en) * | 2013-07-31 | 2013-11-13 | 浙江大学 | Full-automatic three-dimensional conversion method aiming at grating architectural plan |
-
2014
- 2014-09-05 CN CN201410450000.1A patent/CN104252715B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101937579A (en) * | 2010-09-20 | 2011-01-05 | 南京大学 | Method for creating three-dimensional surface model by using perspective sketch |
| CN102880868A (en) * | 2012-08-06 | 2013-01-16 | 上海中和软件有限公司 | Engineering drawing vector conversion and primitive semantic extraction method |
| CN103390088A (en) * | 2013-07-31 | 2013-11-13 | 浙江大学 | Full-automatic three-dimensional conversion method aiming at grating architectural plan |
Non-Patent Citations (6)
| Title |
|---|
| Example-Based 3D object reconstruction from line Drawings;Xue Tianfan et al;《2012 IEEE Conference on computer vision and pattern recognition(CVPR)》;20120616;第303-306页第2-3节 * |
| Separation of line Drawings Based on split faces for 3D objects reconstruction;Changqing Zou et al;《2014 IEEE CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION》;20140623;第692-699页 * |
| 基于单幅线条图的三维立体重建方法综述;郑金鑫;《计算机科学》;20140131;第41卷(第1期);第43-47页 * |
| 工程图纸扫描图像矢量化方法的设计与实现;裘尧波;《中国优秀硕士学位论文全文数据库 信息科技辑》;20140115(第1期);第I138-2026页 * |
| 栅格图像矢量化研究与实现;刘滢滢;《中国优秀硕士学位论文全文数据库 信息科技辑》;20120315(第3期);第I138-2422页 * |
| 线条画矢量化研究;孙玉红 等;《电子技术》;20120731(第7期);第32-34页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN104252715A (en) | 2014-12-31 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Zheng et al. | Non-local scan consolidation for 3D urban scenes | |
| He et al. | Towards fast and accurate real-world depth super-resolution: Benchmark dataset and baseline | |
| CN110443892B (en) | Three-dimensional grid model generation method and device based on single image | |
| Dey et al. | Defining and computing curve-skeletons with medial geodesic function | |
| Li et al. | DNF-Net: A deep normal filtering network for mesh denoising | |
| Hu et al. | Robust hair capture using simulated examples | |
| Song et al. | Deeply supervised depth map super-resolution as novel view synthesis | |
| CN109902583B (en) | Skeleton gesture recognition method based on bidirectional independent circulation neural network | |
| CN109410146A (en) | A kind of image deblurring algorithm based on Bi-Skip-Net | |
| CN109543548A (en) | A kind of face identification method, device and storage medium | |
| CN104252715B (en) | Single line image-based three-dimensional reconstruction method | |
| CN109766866B (en) | Face characteristic point real-time detection method and detection system based on three-dimensional reconstruction | |
| CN103279933A (en) | Method for reconstructing single-image super-resolution based on double-layer model | |
| EP3474185B1 (en) | Classification of 2d images according to types of 3d arrangement | |
| CN112785526A (en) | Three-dimensional point cloud repairing method for graphic processing | |
| Xu et al. | Convolutional neural network for 3d object recognition using volumetric representation | |
| Patrangenaru et al. | Challenges in topological object data analysis | |
| Ahmad et al. | 3D capsule networks for object classification from 3D model data | |
| Sharma et al. | Point cloud upsampling and normal estimation using deep learning for robust surface reconstruction | |
| Demir et al. | Proceduralization of buildings at city scale | |
| Jin et al. | Learning facial expressions with 3D mesh convolutional neural network | |
| Mastan et al. | DeepCFL: Deep contextual features learning from a single image | |
| CN114782417A (en) | Real-time detection method for digital twin characteristics of fan based on edge enhanced image segmentation | |
| Lalos et al. | Signal processing on static and dynamic 3d meshes: Sparse representations and applications | |
| CN111191729B (en) | Three-dimensional object fusion feature representation method based on multi-modal feature fusion |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |