CN104850633A - Three-dimensional model retrieval system and method based on parts division of hand-drawn draft - Google Patents
Three-dimensional model retrieval system and method based on parts division of hand-drawn draft Download PDFInfo
- Publication number
- CN104850633A CN104850633A CN201510267873.3A CN201510267873A CN104850633A CN 104850633 A CN104850633 A CN 104850633A CN 201510267873 A CN201510267873 A CN 201510267873A CN 104850633 A CN104850633 A CN 104850633A
- Authority
- CN
- China
- Prior art keywords
- sampled point
- parts
- image
- carried out
- component tag
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 95
- 238000005070 sampling Methods 0.000 claims abstract description 39
- 239000013598 vector Substances 0.000 claims abstract description 33
- 238000012549 training Methods 0.000 claims abstract description 26
- 238000012545 processing Methods 0.000 claims abstract description 12
- 230000008569 process Effects 0.000 claims description 62
- 230000011218 segmentation Effects 0.000 claims description 51
- 238000000605 extraction Methods 0.000 claims description 42
- 238000006243 chemical reaction Methods 0.000 claims description 25
- 238000005429 filling process Methods 0.000 claims description 16
- 238000011524 similarity measure Methods 0.000 claims description 9
- 239000000284 extract Substances 0.000 claims description 6
- 238000005520 cutting process Methods 0.000 claims description 3
- 238000012163 sequencing technique Methods 0.000 abstract 4
- 238000007781 pre-processing Methods 0.000 abstract 2
- 238000010586 diagram Methods 0.000 description 11
- 230000006870 function Effects 0.000 description 5
- 239000011159 matrix material Substances 0.000 description 5
- 230000000007 visual effect Effects 0.000 description 5
- 241000238631 Hexapoda Species 0.000 description 3
- 238000013461 design Methods 0.000 description 3
- 238000002474 experimental method Methods 0.000 description 3
- 238000011960 computer-aided design Methods 0.000 description 2
- 238000009826 distribution Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 230000002452 interceptive effect Effects 0.000 description 2
- 238000002372 labelling Methods 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 239000000203 mixture Substances 0.000 description 2
- 102000004169 proteins and genes Human genes 0.000 description 2
- 108090000623 proteins and genes Proteins 0.000 description 2
- 230000001174 ascending effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000008878 coupling Effects 0.000 description 1
- 238000010168 coupling process Methods 0.000 description 1
- 238000005859 coupling reaction Methods 0.000 description 1
- 230000003247 decreasing effect Effects 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 230000000977 initiatory effect Effects 0.000 description 1
- 238000003064 k means clustering Methods 0.000 description 1
- 238000002386 leaching Methods 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000010422 painting Methods 0.000 description 1
- 238000005192 partition Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 238000003860 storage Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/951—Indexing; Web crawling techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/58—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Databases & Information Systems (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Library & Information Science (AREA)
- Image Analysis (AREA)
Abstract
The present invention discloses a three-dimensional model retrieval system and method based on parts division of a hand-drawn draft. The system comprises: a preprocessing module, a part marking module, a sampling point feature extracting module, a parts division module, and a similarity calculation and total score sequencing module, wherein the preprocessing module is used for denoising a hand-drawn query draft to obtain a grayscale image, and performing binarization processing, boundary extension processing and image hole filling processing on the grayscale image to obtain a processed image; the part marking module is used for performing equal-interval sampling on the processed image, and adding part tags at sampling points; the sampling point feature extracting module is used for extracting various feature vectors of the sampling points; the parts division module is used for performing division model training according to the various feature vectors of the sample points to which the part tags have been added; and the similarity calculation and total score sequencing module is used for performing local similarity calculation on parts, performing sequencing according to total scores, and feeding back a sequencing result to a client. Implementation of the embodiments of the present invention can make three-dimensional model retrieval based on the hand-drawn draft more accurate and effective.
Description
Technical field
The present invention relates to computer image processing technology field, particularly relate to a kind of three-dimensional model searching system based on the segmentation of cartographical sketching parts and method.
Background technology
In recent years, along with the fast development in the fields such as computer-aided design (CAD), computer-aided manufacturing, virtual reality, three-dimensional animation and 3d gaming, the three-dimensional model quantity sharp increase on internet.But three-dimensional model is different from the multimedia messagess such as traditional picture, audio or video, itself contain many detailed information be difficult to literal expression out.
But current method for searching three-dimension model is still not fully up to expectations in application.On the one hand, when user needs certain three-dimensional model resource, often not ready-made on hand model file; On the other hand, universal fast along with touch-screen and electronic pen, the mode that user can easily pass cartographical sketching sketches the contours of the profile of model.The cartographical sketching of three-dimensional model, can be considered as is the outline line from certain visual angle projection view.Cartographical sketching can be simple outer contour, also can comprise the detailed information of inner outline.Because user's art activities of cartographical sketching is different, input equipment is different, and the level of detail describing model is also not quite similar naturally.And the cartographical sketching of three-dimensional model comprises overlapping, be separated or inc component outline line usually; existing correlative study is usually based on the craft segmentation carried out cartographical sketching or mark; although these manual information of specifying contribute to computing machine and analyze cartographical sketching; but it requires to follow certain rule constrain during user's cartographical sketching usually; this limits the degrees of freedom of user's Freehandhand-drawing to a certain extent, proposes requirement in other words to the foundation of painting of user.
Classify according to retrieval mode, the retrieval of current three-dimensional model is mainly divided into two large classes, is traditional text based retrieval (Text-based Retrieval) method and content-based retrieval (Content-based Retrieval) method respectively.
(1) text based method for searching three-dimension model
Text based method for searching three-dimension model, based on keyword, is retrieval mode the most general at present.This needs the keyword adding to describe it to the three-dimensional model in database artificially, the 3D protein retrieval system (3D Protein Retrieval System) etc. of the 3D model bank (3D Warehouse) of such as SketchUp, the special model library of TurboSquid and TaiWan, China university, the model index platform of the commercialization can some being found large-scale on the net now, they are this kind of three-dimensional model search mode based on keyword mostly.
(2) content-based method for searching three-dimension model
Content-based retrieval method is the study hotspot of three-dimensional model search.Be illustrated in figure 1 the basic framework of content-based three-dimensional model search, framework entirety is divided into off-line part and online part.For off-line part, each 3D model needs to indicate with shape description symbols.In order to promote recall precision, usually index is set up to aspect of model descriptor each in database.For online part, the input carrying out query express mainly can be divided into two kinds of modes: be a kind ofly to provide a three-dimensional model similar with object module; Another kind is the sketch of Freehandhand-drawing object module.After calculating feature descriptor, user search is inputted aspect of model descriptor in the descriptor of data and database and carry out similarity-rough set, result returns by the order of then successively decreasing according to similarity size, and presents to user visually.
The shortcoming existed in prior art:
(1) text based method for searching three-dimension model
Traditional mode based on text key word can not be useful in the scene of three-dimensional model search well, its main cause has 3 points: first, three-dimensional model has complicated topological structure, shape facility, and of a great variety, itself contain a lot of detailed information and be difficult to express clearly with several keyword.The second, the process need that tags adding text key word to three-dimensional model completes by hand, and on internet, three-dimensional model quantity increases fast, and the mode manually added is comparatively loaded down with trivial details, and workload is also very large.3rd, because different people is different to the understanding of miscellaneous three-dimensional model, its keyword of the description expected also has larger difference, easily cause the label of search key and object module inconsistent, and the manual mode adding keyword label is limited to markup language kind, is also not easy to carry out internationalization and promotes.Just for these reasons, only adopt simple keyword to retrieve, success ratio can be very low, can not get the result wanted time many.Such as, user wants the car retrieving certain given configuration and pattern, so only relies on the result that keyword is difficult to search accurately, is satisfied with.
(2) content-based method for searching three-dimension model
For the model index based on three-dimensional model example, its shortcoming is that user is when initiating retrieval, usually be difficult to find a most suitable model instance as input, if because user has most suitable object module at hand, so also just need not retrieved.
For the three-dimensional model search based on cartographical sketching, its shortcoming is that it does not consider the structure of sketch entirety usually, and it is just considered based on local, region; Its another shortcoming is exactly more responsive to the style of user's sketch, if the outline line of user in local draws stylistic differences comparatively greatly, the result difference that so its extracts will amplify, and this will certainly affect final result for retrieval.
Summary of the invention
The object of the invention is to overcome the deficiencies in the prior art, the invention provides a kind of three-dimensional model searching system based on the segmentation of cartographical sketching parts and method, the three-dimensional model search based on cartographical sketching can be made more precisely effective.
In order to solve the problem, the present invention proposes a kind of three-dimensional model searching system based on the segmentation of cartographical sketching parts, described system comprises:
Pretreatment module, for receiving Freehandhand-drawing inquiry sketch, carrying out denoising to described Freehandhand-drawing inquiry sketch and obtaining gray-scale map, and carry out binary conversion treatment, border extension process, vacancy filling process to described gray-scale map, obtain the image after process;
Parts mark module, for carrying out equal interval sampling to the image after described process, obtains sampled point, and adds component tag to described sampled point;
Sampled point characteristic extracting module, for extracting each feature vectors of described sampled point;
Parts segmentation module, for carrying out parted pattern training according to each feature vectors of the sampled point after interpolation component tag;
Similarity Measure and overall score order module, for carrying out parts local shape factor and the calculating of parts local similarity based on parted pattern, and the extraction of view global characteristics and view overall situation Similarity Measure are carried out to the image after described process, sort according to overall score, and ranking results is returned to client.
Preferably, described pretreatment module comprises:
Sketch denoising unit, obtains gray-scale map for carrying out denoising to described Freehandhand-drawing inquiry sketch;
Binary conversion treatment unit, for carrying out binary conversion treatment to described gray-scale map;
Border extension processing unit, fills process for carrying out blank to the image surrounding after binary conversion treatment;
Vacancy fills processing unit, carries out vacancy filling process for the image after filling process to blank.
Preferably, described parts mark module comprises:
Outline line extraction unit, for carrying out outline line extraction to the image after described process;
Sampling unit, for carrying out equal interval sampling to the image after Extracting contour, obtains sampled point;
Parts indexing unit, for adding component tag to described sampled point.
Preferably, described sampled point characteristic extracting module comprises:
Unitary feature extraction unit, for carrying out unitary feature extraction to the sampled point after interpolation component tag;
Binary feature extraction unit, for carrying out binary feature extraction to the sampled point after interpolation component tag.
Preferably, described parts segmentation module comprises:
Parted pattern training unit, for carrying out parted pattern training according to each feature vectors of the sampled point after interpolation component tag;
Parts cutting unit, for carrying out parts segmentation according to parted pattern to the sampled point after interpolation component tag.
Correspondingly, the present invention also provides a kind of method for searching three-dimension model based on the segmentation of cartographical sketching parts, and described method comprises:
Receive Freehandhand-drawing inquiry sketch, denoising is carried out to described Freehandhand-drawing inquiry sketch and obtains gray-scale map, and binary conversion treatment, border extension process, vacancy filling process are carried out to described gray-scale map, obtain the image after process;
Equal interval sampling is carried out to the image after described process, obtains sampled point, and component tag is added to described sampled point;
Extract each feature vectors of described sampled point;
Each feature vectors according to the sampled point after interpolation component tag carries out parted pattern training;
Parts local shape factor and the calculating of parts local similarity is carried out based on parted pattern, and the extraction of view global characteristics and view overall situation Similarity Measure are carried out to the image after described process, sort according to overall score, and ranking results is returned to client.
Preferably, describedly denoising carried out to described Freehandhand-drawing inquiry sketch obtain gray-scale map, and binary conversion treatment is carried out to described gray-scale map, border extension process, vacancy fill process, obtain the step of the image after process, comprising:
Denoising is carried out to described Freehandhand-drawing inquiry sketch and obtains gray-scale map;
Binary conversion treatment is carried out to described gray-scale map;
Blank is carried out to the image surrounding after binary conversion treatment and fills process;
Image after filling process to blank carries out vacancy and fills process.
Preferably, described equal interval sampling is carried out to the image after described process, obtains sampled point, and the step of component tag is added to described sampled point, comprising:
Outline line extraction is carried out to the image after described process;
Equal interval sampling is carried out to the image after Extracting contour, obtains sampled point;
Component tag is added to described sampled point.
Preferably, the step of each feature vectors of the described sampled point of described extraction, comprising:
Unitary feature extraction is carried out to the sampled point after adding component tag;
Binary feature extraction is carried out to the sampled point after adding component tag.
Preferably, described each feature vectors according to the sampled point after interpolation component tag carries out the step of parted pattern training, comprising:
Each feature vectors according to the sampled point after interpolation component tag carries out parted pattern training;
According to parted pattern, parts segmentation is carried out to the sampled point after interpolation component tag.
In embodiments of the present invention, topology information between the geological information of comprehensive utilization cartographical sketching parts, parts and the global information of view picture view, and be provided with the mechanism of three-view diagram dynamic weight index, amplify the impact of important visual angle in overall score, thus make the three-dimensional model search based on cartographical sketching more precisely effective; In addition, can in the application scenarios separately for the segmentation of the Freehandhand-drawing figure parts such as sketch understanding, sketch classification.
Accompanying drawing explanation
In order to be illustrated more clearly in the embodiment of the present invention or technical scheme of the prior art, be briefly described to the accompanying drawing used required in embodiment or description of the prior art below, apparently, accompanying drawing in the following describes is only some embodiments of the present invention, for those of ordinary skill in the art, under the prerequisite not paying creative work, other accompanying drawing can also be obtained according to these accompanying drawings.
Fig. 1 is the basic framework schematic diagram of three-dimensional model search content-based in prior art;
Fig. 2 is the structure composition schematic diagram of the three-dimensional model searching system based on the segmentation of cartographical sketching parts of the embodiment of the present invention;
Fig. 3 is the internal processes schematic diagram of the three-dimensional model searching system based on the segmentation of cartographical sketching parts of the embodiment of the present invention;
Fig. 4 adds tagged effect schematic diagram in the embodiment of the present invention;
Fig. 5 is the schematic flow sheet of the method for searching three-dimension model based on the segmentation of cartographical sketching parts of the embodiment of the present invention.
Embodiment
Below in conjunction with the accompanying drawing in the embodiment of the present invention, be clearly and completely described the technical scheme in the embodiment of the present invention, obviously, described embodiment is only the present invention's part embodiment, instead of whole embodiments.Based on the embodiment in the present invention, those of ordinary skill in the art, not making the every other embodiment obtained under creative work prerequisite, belong to the scope of protection of the invention.
Fig. 2 is the structure composition schematic diagram of the three-dimensional model searching system based on the segmentation of cartographical sketching parts of the embodiment of the present invention, and as shown in Figure 2, this system comprises:
Pretreatment module 1, for receiving Freehandhand-drawing inquiry sketch, carrying out denoising to Freehandhand-drawing inquiry sketch and obtaining gray-scale map, and carries out binary conversion treatment, border extension process, vacancy filling process to gray-scale map, obtains the image after process;
Parts mark module 2, carries out equal interval sampling for the image after processing pretreatment module 1, obtains sampled point, and add component tag to sampled point;
Sampled point characteristic extracting module 3, for each feature vectors of the sampled point that extracting parts mark module 2 obtains;
Parts segmentation module 4, for carrying out parted pattern training according to each feature vectors of the sampled point after interpolation component tag;
Similarity Measure and overall score order module 5, for carrying out parts local shape factor and the calculating of parts local similarity based on parted pattern, and the extraction of view global characteristics and view overall situation Similarity Measure are carried out to the image after process, sort according to overall score, and ranking results is returned to client.
Fig. 3 is the internal processes of the three-dimensional model searching system based on the segmentation of cartographical sketching parts of the embodiment of the present invention, is described in detail to the system of the embodiment of the present invention below in conjunction with Fig. 2, Fig. 3.
The cartographical sketching of user's input when retrieving is different from the projection view of three-dimensional model.Because user is by mouse, touch-screen or pointer skeletonizing, cartographical sketching is often accurate not, and its outline line often comprises noise, therefore needs to carry out necessary pre-service to user's cartographical sketching, just can carry out ensuing handling procedure.
Wherein, pretreatment module 1 comprises:
Sketch denoising unit, obtains gray-scale map for carrying out denoising to Freehandhand-drawing inquiry sketch;
Binary conversion treatment unit, for carrying out binary conversion treatment to gray-scale map;
Border extension processing unit, fills process for carrying out blank to the image surrounding after binary conversion treatment;
Vacancy fills processing unit, carries out vacancy filling process for the image after filling process to blank.
Sketch denoising: can have an impact to following feature extraction because cartographical sketching end is not closed, pre-service is carried out in the embodiment of the present invention, the starting point of every unicursal and terminal point coordinate when recording user is drawn, if the Euclidean distance between the extreme coordinates of the extreme coordinates of certain unicursal and other strokes is less than the threshold value of 30 pixels, so just line between these two points.
Bounding box process: because the drafting part of different sketch is not of uniform size, in order to effectively, as one man carry out follow-up feature extraction and machine learning, needs to carry out bounding box process to view, obtains the regional area having sketch lines in original image.
Keeping proportional zoom process: in order to meet graphical rule unchangeability, allow the blank border of all figure all minimize as much as possible, needing to carry out the process of maintenance proportional zoom.With 180 divided by the length of bounding box longest edge as the scale factor of convergent-divergent, thus carry out image scaling.
Binary conversion treatment: cartographical sketching has become gray-scale map after scaling, needs to carry out binary conversion treatment, so that the global characteristics in model index flow process extracts.
Border extension process: in order to make sketch drafting region placed in the middle, needing to carry out border extension process, image surrounding being filled in the blanks, making image size unified is 200*200 pixel.
Vacancy fills process: the outline line due to sketch comprises outer contour and inner outline, for model index scene, usual outer contour enough distinguishes different classes of model, think cartographical sketching compared to projection view under similar details represents degree situation, need first to carry out filling pre-service to it.
Further, parts mark module 2 comprises:
Outline line extraction unit, for carrying out outline line extraction to the image after process;
Sampling unit, for carrying out equal interval sampling to the image after Extracting contour, obtains sampled point;
Parts indexing unit, for adding component tag to sampled point.
(1) large class divides
Model classification in database is numerous, and topological structure varies, and it is necessary for therefore before parts mark, carrying out the division of large class.The standard that large class divides is two aspects: on the one hand, the model of similar topological structure as much as possible should be made to be included into same large class, carry out the training study of parts segmentation together; On the other hand, the large class number that all model partitions in database should be made to go out is the least possible, thus reduces the total degree of online part cartographical sketching by each large class segmentation.
(2) based on equally spaced sampled point design and component tag
Consider the follow-up feature extraction to each sampled point, require that sampled point size must ensure following three aspects: sampled point is enough intensive, in order to avoid different parts fall into same sampled point, causes the inaccurate of component tag; Sampled point should not be too intensive, in order to avoid extract total computing high cost of feature, causing when retrieving cannot real-time response; The size of sampled point determines to consider the sampled point size of multiple unitary feature when extracting, the relation in integral multiple between the sampled point length of side making to extract different characteristic, thus ensure that each sampled point can obtain reasonably, effective Feature Combination.
Based on these reasons above-mentioned, attempt through great many of experiments, the final mode adopting equal interval sampling, to every width outline line along outline line apart from every 10 pixel extraction 1 sampled point.Component tag is added to sampled point, as head, body, four limbs, tail etc., for doing parts segmentation to the view of large class model, trains by the local feature vectors of each view samples point in current large class.
(3) based on the automatic pre-segmentation of discrete curve evolvement model and framework information
Carrying out in the research of parts segmentation based on view, one intuitively idea be exactly analyze based on the geometric properties of view outline line, thus determine the cut-off rule between adjacent component, and then carry out the segmentation of parts.Adding tagged efficiency to improve, in embodiments of the present invention, adopting the view parts auto Segmentation mode based on discrete curve evolvement model and skeleton, automatically carrying out the pre-segmentation of parts.
By the Extreme points set of the simplified polygon of discrete curve evolvement model and the difference set of skeleton distal point set, and on be point of contact with each element in this set, take axis as other all point of contacts of the incircle in the center of circle, thus form pre-segmentation point set.Found through experiments, this set is all the set that the potential cut-point of most models is formed, therefore adopt this set as outline line segmentation foundation, mark is added according to outline line sampled point sequential segment, eliminate the process that each parts circle selects polygonal embracing cartridge, thus effectively promote labeling effciency.
(4) design of parts mark small tool
Conveniently to sampling number according to interpolation component tag, have employed and interactive add and revise the mode marked.Be add tagged schematic diagram as shown in Figure 4, adopt mouse to click the mode of delimiting parts polygonal embracing cartridge, can mark the sampled point at parts place easily.
The label of large class label and model class is not added in output file, mainly consider in later stage convenient constantly adjustment feature extraction mode and parameter, as long as and the processing mode of bounding box and sampled point is constant, the component tag of sampled point is not just by the impact of adjustment feature extracting method.Therefore reduce the degree of coupling of label and characteristic extraction part as far as possible, these two parts are peeled away, in accordance with unified file designation rule, write respectively in two files under same catalogue, facilitate post-processed.Tagged work fully completes in interactive small tool, does not interfere with each other with automatic characteristic extraction procedure.
Further, sampled point characteristic extracting module 3 comprises:
Unitary feature extraction unit, for carrying out unitary feature extraction to the sampled point after interpolation component tag;
Binary feature extraction unit, for carrying out binary feature extraction to the sampled point after interpolation component tag.
(1) the unitary feature of single sampled point self
Unitary feature is for characterizing the feature of each sampled point inside.The unitary feature adopted in the embodiment of the present invention, all based on sampled point, calculates each feature vectors of each sampled point.The details of eigen leaching process will be set forth respectively below.
2D shape characteristics of diameters: to the calculating of each sampled point angle according to line between itself and neighbouring sample point, calculate the tangential direction of this point, then along the direction vertical with this tangent line, ray is sent in image mask, meet at the marginal point of image opposite side, calculate the length at shaped interior ray portion.Similarly, calculate the marginal point meeting at image opposite side with the ray that vertical line both sides become the direction of 30 °, 60 ° angles to draw, computational length value too, finally asks for the mean value of these distance values, as the 2D shape characteristics of diameters of this sample point, and totally 1 dimension.
Sampled point is to the distance feature of image center: using the part of the Euclidean distance between sampled point to image center as unitary feature.
Average Euclidean distance feature: the average euclidean distance metric based on sampled point, be used for characterizing each sampled point apart from other sampled point away from degree.Such as in the view of a width insect, the average Euclidean distance of insect leg is usually farther than other parts.The average Euclidean distance of each sampled point is tried to achieve by the distance matrix of SC sampled point, if having multiple sampled point in each sampled point, adopts sample point Euclidean distance mean value.Euclidean distance obtains to the Euclidean distance mean value of other each sample point by calculating each sample point.Calculate simultaneously average square and the 10th, the 20th, the 30th until the data of the 90th quantile, then by these 11 statistical measures unifications divided by the maximal value in all sampled point Euclidean distances of present image, thus to be normalized, to form 11 dimensional vectors.
Shape context histogram feature: Shape context algorithm equally spaced gets sampled point on object edge line, calculates each sampled point relative to the Euclidean distance of other each sampled point and angle.
Place connected component proportion feature: because some Freehandhand-drawing figure and view are made up of multiple outline line, each profile line segment characterizes semantically independently parts usually.Adopt 1 dimensional vector in the embodiment of the present invention, the connected component for characterizing current sampling point place accounts for the ratio of entire image.First, at Extracting contour and the stage of down-sampling record the sampling number of each section of outline line respectively, then the ratio of total sampling number is accounted for by the sampling number of the outline line judging current sampling point place, thus draw a ratio characteristic had nothing to do with sampling step length, in order to characterize the proportion of outline line in total outline line at current sampling point place.Find in the experiment of insects, normally leg, feeler etc. are in the comparatively long and narrow parts of object periphery to have the sampled point of less connected component proportion feature.
(2) binary feature between adjacent sampled point
Binary feature weighs tag compliance between each sampled point and adjacent sampled point.Therefore, before calculating binary feature, need first to obtain each sampled point abutment points information on outline line.Because sampled point sequence is orderly, so syntople by whether being greater than sampling step length threshold value to judge to the Euclidean distance of consecutive point in sampled point sequence, if be greater than threshold value, can not being abutment points, otherwise being then abutment points.In addition, consider that some models have many profile line segments, so also need to judge whether adjoin between the starting sample point of every section of outline line and termination sampled point.Binary feature needs enough discriminations, namely requires that the binary feature of the binary feature of parts intersection sampled point and non-parts intersection sampled point should have larger difference.The absolute value of difference of 2D shape diameter and the absolute value of the difference of tangential direction, as binary feature.The numerical value of each component value of binary feature vector of parts intersection sampled point is larger; And each component values of binary feature of sampled point not near parts point of interface is less, therefore there is significant discrimination to the intersection of parts.
Further, parts segmentation module 4 comprises:
Parted pattern training unit, for carrying out parted pattern training according to each feature vectors of the sampled point after interpolation component tag;
Parts cutting unit, for carrying out parts segmentation according to parted pattern to the sampled point after interpolation component tag.
In concrete enforcement, after interpolation component tag, adopt condition random field (CRF) model, carry out the training of parted pattern.
(1) conditional random field models
The objective energy function of CRF model is made up of unitary item and binary item.Wherein unitary item weighs the unitary feature of sampled point and the consistance of label thereof; And the label that binary Xiang Ze is weighed between sampled point and abutment points by binary feature is compatible.
Here adopt the parts segmentation based on CRF model and labeling method, be respectively used to the training study of the parts segmentation of each large class internal model view.Namely the optimum label calculating all sampled points requires to minimize objective function, as shown in formula (1):
Objective function comprises two large divisions, wherein E
1for unitary energy term, E
2for dual-energy item.
Unitary energy term E
1: in order to assess a sorter.Sorter, using the proper vector x of sampled point as input, exports under this characteristic condition, probability distribution P (c|x, the θ of sampled point label
1).JointBoost sorter is adopted to carry out machine learning.As formula (2) is depicted as the account form of unitary energy term:
E
1(c；x,θ
1)＝-logP(c|x,θ
1) (2)
In formula (2), x is unitary feature, and the unitary energy term of each label c equals under this proper vector condition, the negative logarithm of sampled point label probability distribution.
Dual-energy item E
2: be used for characterizing each sampled point and probability outline line being labeled as different label between neighbouring sample point, its definition is as shown in formula (3):
E
2(c,c'；y,θ
2)＝W(c,c')·[-κlogP(c≠c'|y,ξ)+μ] (3)
In formula (3), y is binary feature.P (c ≠ c'|y, ξ) characterizes the different possibility size of label, and it is the function of a geometric properties.Label punishment matrix W (c, c') represents the degree of compatibility between label c and c'.It is symmetric matrix, and in matrix, each element is both initialized to 9999, and learning process obtains the penalty value between often pair of label iteratively.
(2) the iteration optimization learning method of conditional random field models parameter
After extracting the unitary of sampled point, binary feature, adopt the optimized mode of iteration to learn the parameter of CRF model, training sampled point set mark is divided into 5 equal portions randomly, wherein 4 parts as training set, and 1 part collects as verifying.First, with sample set training JointBoost sorter, the partial parameters of study unitary item and binary item.Then, collect Optimized Segmentation result in an iterative manner with checking, thus remaining parameter in study CRF model binary item.
The parts that each model three projection views are partitioned into being carried out respectively the training of sorter, is also that three views calculate the votes dropped in respective classes respectively when retrieving.Such ratio precision of carrying out training of all being put together by view component feature is higher, and reduces the calculated amount of online part, thus improves recall precision.
Adopt parts segmentation can obtain sampled point parts flag sequence.The component tag of each sampled point of this sequential recording and coordinate position in the picture thereof.
Because sampled point mark result is along outline line ordered arrangement, the embodiment of the present invention is based on this feature design part figure generating algorithm, read in each sampled point mark and image coordinate point thereof in turn, record a upper sample point coordinate position of often kind of component tag, and calculate the Euclidean distance between adjacent same label sampled point based on this, if distance is less than the equidistant sampling step length of outline line, then at this point-to-point transmission connecting line segment.If read in the new component tag of non-generating unit figure, corresponding this component diagram matrix newly-built.Also all parts sampling number is recorded while reading in sampled point mark.Carry out this process iteratively, until sampled point flag sequence has read.Finally, bounding box and extended boundary are calculated respectively to the image of each parts, make parts placed in the middle, thus generate the component diagram being of a size of 100*100 pixel.The weight of each parts is the ratio that this parts sampling number accounts for total sampling number.
The Zernike moment characteristics descriptor adopted, forms the global characteristics of view.Zernike square meets rotational invariance, has good discrimination to difformity profile.Zernike square has two parameters, and namely n and m, n represent the rank of Zernike square, and m represents the repeat number of Zernike square.The combination often organizing n and m can obtain a plural Zernike square value, adopts the component of amplitude as global characteristics vector of Zernike square.The more results of component that this feature descriptor is selected are more accurate, but the more calculating of component is consuming time also increases thereupon.Take into account retrieval accuracy and calculate two aspects consuming time, selecting the Zernike square of 10 dimensions, the combination of itself n and m value is as shown in table 1.
N and the m value combination of table 1 Zernike square global characteristics
| N value | 3 | 5 | 7 | 9 | 11 | 4 | 6 | 8 | 10 | 12 |
| M value | 3 | 3 | 3 | 3 | 3 | 4 | 4 | 4 | 4 | 4 |
Laterally, longitudinally 32 parts are divided into image, successively with 200/32 pixel for step-length movable block.In the process of movable block, if run into the situation without any outline line in block, then directly abandon the statistics to this block, thus avoid and that cause characteristic component blank due to image local be in a large number 0 situation.
In off-line part, with K-means clustering algorithm, cluster is carried out to Gabor characteristic, form 128 bunches, each bunch of center is preserved, form dictionary.Finally, the number of each Gabor characteristic vector that statistical distance each bunch of center is nearest.
The direct syntople of component tag provides the topological structure between parts, and topological features has very strong robustness for the torsion of parts.Such as, the leg of a horse of running and the geometric properties of the leg of horse of standing differ greatly, no matter but whether be the state of running, leg is all adjacent with body part.Set up the graph model of topological structure between parts based on this point characteristic, using the node of parts as figure, fillet between adjacent component, realizes with adjacency list.If these parts of cartographical sketching are different from the adjacent part of projection view corresponding component, then the difference of two figure topology aspects punished, namely reduce similarity score.
Consider topology information between bottom geological information and parts, account for the ratio of the total sampling number of all parts according to this parts sampling number, and the weighted value of these parts of ratio-dependent that these parts occur in all model banies.Formula (4) is depicted as the computing formula of parts weight:
Formula (5) is depicted as the computing formula of the similarity score at view visual angle between sketch and model:
The parts topological structure of sketch may not meet with the topological structure of projection view, adopts rational penalty term here, thus has considered topological structure difference.Compare in topological structure adjacency list between corresponding component, if there is the situation that topological structure is inconsistent, then the weight according to these parts of projection view is punished similarity score.
The quantity of information that the projection view of three-dimensional model different visual angles provides is different.Such as, almost cannot from the top view of people obtaining parts information, also cannot find out that it belongs to the model projection of people, the semantic parts such as head, body, four limbs that the front elevation of people then clearly represents.The component count that only view publishing goes out reflects the quantity of information that projection view can provide objectively.Based on this feature, as shown in formula (6), calculate the component count be partitioned in front elevation, side view and vertical view respectively account for the ratio that three-view diagram is partitioned into total parts count, respectively as front elevation weight w
i, front, side view weight w
i, sidewith vertical view weight w
i, top, the computing formula of each model overall similarity scoring in current large class is depicted as with formula (7):
Finally, according to overall score ascending sort, front 200 models are returned to user, and corresponding thumbnail presents in a browser by paging, namely completes primary retrieval process.
Correspondingly, the embodiment of the present invention also provides a kind of method for searching three-dimension model based on the segmentation of cartographical sketching parts, and as shown in Figure 5, the method comprises:
S501, receives Freehandhand-drawing inquiry sketch, carries out denoising obtain gray-scale map to Freehandhand-drawing inquiry sketch, and carries out binary conversion treatment, border extension process, vacancy filling process to gray-scale map, obtains the image after process;
S502, carries out equal interval sampling to the image after process, obtains sampled point, and adds component tag to sampled point;
S503, extracts each feature vectors of sampled point;
S504, each feature vectors according to the sampled point after interpolation component tag carries out parted pattern training;
S505, parts local shape factor and the calculating of parts local similarity is carried out based on parted pattern, and the extraction of view global characteristics and view overall situation Similarity Measure are carried out to the image after process, sort according to overall score, and ranking results is returned to client.
Wherein, S501 comprises further:
Denoising is carried out to Freehandhand-drawing inquiry sketch and obtains gray-scale map;
Binary conversion treatment is carried out to gray-scale map;
Blank is carried out to the image surrounding after binary conversion treatment and fills process;
Image after filling process to blank carries out vacancy and fills process.
S502 comprises further:
Outline line extraction is carried out to the image after process;
Equal interval sampling is carried out to the image after Extracting contour, obtains sampled point;
Component tag is added to sampled point.
S503 comprises further:
Unitary feature extraction is carried out to the sampled point after adding component tag;
Binary feature extraction is carried out to the sampled point after adding component tag.
S504 comprises further:
Each feature vectors according to the sampled point after interpolation component tag carries out parted pattern training;
According to parted pattern, parts segmentation is carried out to the sampled point after interpolation component tag.
Particularly, the implementation procedure of the inventive method embodiment see the associated description of the principle of work of system related functions module, can repeat no more here.
In embodiments of the present invention, topology information between the geological information of comprehensive utilization cartographical sketching parts, parts and the global information of view picture view, and be provided with the mechanism of three-view diagram dynamic weight index, amplify the impact of important visual angle in overall score, thus make the three-dimensional model search based on cartographical sketching more precisely effective; In addition, can in the application scenarios separately for the segmentation of the Freehandhand-drawing figure parts such as sketch understanding, sketch classification.
One of ordinary skill in the art will appreciate that all or part of step in the various methods of above-described embodiment is that the hardware that can carry out instruction relevant by program has come, this program can be stored in a computer-readable recording medium, storage medium can comprise: ROM (read-only memory) (ROM, Read OnlyMemory), random access memory (RAM, Random Access Memory), disk or CD etc.
In addition, above to the embodiment of the present invention provide based on cartographical sketching parts segmentation three-dimensional model searching system and method be described in detail, apply specific case herein to set forth principle of the present invention and embodiment, the explanation of above embodiment just understands method of the present invention and core concept thereof for helping; Meanwhile, for one of ordinary skill in the art, according to thought of the present invention, all will change in specific embodiments and applications, in sum, this description should not be construed as limitation of the present invention.
Claims (10)
1., based on a three-dimensional model searching system for cartographical sketching parts segmentation, it is characterized in that, described system comprises:
Pretreatment module, for receiving Freehandhand-drawing inquiry sketch, carrying out denoising to described Freehandhand-drawing inquiry sketch and obtaining gray-scale map, and carry out binary conversion treatment, border extension process, vacancy filling process to described gray-scale map, obtain the image after process;
Parts mark module, for carrying out equal interval sampling to the image after described process, obtains sampled point, and adds component tag to described sampled point;
Sampled point characteristic extracting module, for extracting each feature vectors of described sampled point;
Parts segmentation module, for carrying out parted pattern training according to each feature vectors of the sampled point after interpolation component tag;
Similarity Measure and overall score order module, for carrying out parts local shape factor and the calculating of parts local similarity based on parted pattern, and the extraction of view global characteristics and view overall situation Similarity Measure are carried out to the image after described process, sort according to overall score, and ranking results is returned to client.
2., as claimed in claim 1 based on the three-dimensional model searching system of cartographical sketching parts segmentation, it is characterized in that, described pretreatment module comprises:
Sketch denoising unit, obtains gray-scale map for carrying out denoising to described Freehandhand-drawing inquiry sketch;
Binary conversion treatment unit, for carrying out binary conversion treatment to described gray-scale map;
Border extension processing unit, fills process for carrying out blank to the image surrounding after binary conversion treatment;
Vacancy fills processing unit, carries out vacancy filling process for the image after filling process to blank.
3., as claimed in claim 1 based on the three-dimensional model searching system of cartographical sketching parts segmentation, it is characterized in that, described parts mark module comprises:
Outline line extraction unit, for carrying out outline line extraction to the image after described process;
Sampling unit, for carrying out equal interval sampling to the image after Extracting contour, obtains sampled point;
Parts indexing unit, for adding component tag to described sampled point.
4., as claimed in claim 1 based on the three-dimensional model searching system of cartographical sketching parts segmentation, it is characterized in that, described sampled point characteristic extracting module comprises:
Unitary feature extraction unit, for carrying out unitary feature extraction to the sampled point after interpolation component tag;
Binary feature extraction unit, for carrying out binary feature extraction to the sampled point after interpolation component tag.
5. as claimed in claim 1 based on the three-dimensional model searching system of cartographical sketching parts segmentation, it is characterized in that, described parts segmentation module comprises:
Parted pattern training unit, for carrying out parted pattern training according to each feature vectors of the sampled point after interpolation component tag;
Parts cutting unit, for carrying out parts segmentation according to parted pattern to the sampled point after interpolation component tag.
6., based on a method for searching three-dimension model for cartographical sketching parts segmentation, it is characterized in that,
Receive Freehandhand-drawing inquiry sketch, denoising is carried out to described Freehandhand-drawing inquiry sketch and obtains gray-scale map, and binary conversion treatment, border extension process, vacancy filling process are carried out to described gray-scale map, obtain the image after process;
Equal interval sampling is carried out to the image after described process, obtains sampled point, and component tag is added to described sampled point;
Extract each feature vectors of described sampled point;
Each feature vectors according to the sampled point after interpolation component tag carries out parted pattern training;
Parts local shape factor and the calculating of parts local similarity is carried out based on parted pattern, and the extraction of view global characteristics and view overall situation Similarity Measure are carried out to the image after described process, sort according to overall score, and ranking results is returned to client.
7. as claimed in claim 6 based on the method for searching three-dimension model of cartographical sketching parts segmentation, it is characterized in that, describedly denoising is carried out to described Freehandhand-drawing inquiry sketch obtain gray-scale map, and binary conversion treatment, border extension process, vacancy filling process are carried out to described gray-scale map, obtain the step of the image after process, comprising:
Denoising is carried out to described Freehandhand-drawing inquiry sketch and obtains gray-scale map;
Binary conversion treatment is carried out to described gray-scale map;
Blank is carried out to the image surrounding after binary conversion treatment and fills process;
Image after filling process to blank carries out vacancy and fills process.
8. as claimed in claim 6 based on the method for searching three-dimension model of cartographical sketching parts segmentation, it is characterized in that, described equal interval sampling carried out to the image after described process, obtain sampled point, and the step of component tag is added to described sampled point, comprising:
Outline line extraction is carried out to the image after described process;
Equal interval sampling is carried out to the image after Extracting contour, obtains sampled point;
Component tag is added to described sampled point.
9., as claimed in claim 6 based on the method for searching three-dimension model of cartographical sketching parts segmentation, it is characterized in that, the step of each feature vectors of the described sampled point of described extraction, comprising:
Unitary feature extraction is carried out to the sampled point after adding component tag;
Binary feature extraction is carried out to the sampled point after adding component tag.
10. as claimed in claim 6 based on the method for searching three-dimension model of cartographical sketching parts segmentation, it is characterized in that, described each feature vectors according to the sampled point after interpolation component tag carries out the step of parted pattern training, comprising:
Each feature vectors according to the sampled point after interpolation component tag carries out parted pattern training;
According to parted pattern, parts segmentation is carried out to the sampled point after interpolation component tag.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510267873.3A CN104850633B (en) | 2015-05-22 | 2015-05-22 | A kind of three-dimensional model searching system and method based on the segmentation of cartographical sketching component |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510267873.3A CN104850633B (en) | 2015-05-22 | 2015-05-22 | A kind of three-dimensional model searching system and method based on the segmentation of cartographical sketching component |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104850633A true CN104850633A (en) | 2015-08-19 |
| CN104850633B CN104850633B (en) | 2018-10-12 |
Family
ID=53850277
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510267873.3A Active CN104850633B (en) | 2015-05-22 | 2015-05-22 | A kind of three-dimensional model searching system and method based on the segmentation of cartographical sketching component |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104850633B (en) |
Cited By (23)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105243137A (en) * | 2015-09-30 | 2016-01-13 | 华南理工大学 | Draft-based three-dimensional model retrieval viewpoint selection method |
| CN106126581A (en) * | 2016-06-20 | 2016-11-16 | 复旦大学 | Cartographical sketching image search method based on degree of depth study |
| CN106530397A (en) * | 2016-10-13 | 2017-03-22 | 成都希盟泰克科技发展有限公司 | Geological surface three-dimensional reconstruction method based on sparse profile geological contours |
| CN106874350A (en) * | 2016-12-27 | 2017-06-20 | 合肥阿巴赛信息科技有限公司 | A kind of diamond ring search method and system based on sketch and distance field |
| CN107122396A (en) * | 2017-03-13 | 2017-09-01 | 西北大学 | Three-dimensional model searching algorithm based on depth convolutional neural networks |
| CN107220277A (en) * | 2017-04-14 | 2017-09-29 | 西北大学 | Image retrieval algorithm based on cartographical sketching |
| CN108109205A (en) * | 2017-12-08 | 2018-06-01 | 山东师范大学 | A kind of incomplete model index and method for reconstructing based on holes filling |
| CN108170823A (en) * | 2018-01-04 | 2018-06-15 | 江西师范大学 | A kind of Freehandhand-drawing interactive three-dimensional model retrieval method understood based on high-level semantic attribute |
| CN109308324A (en) * | 2018-09-08 | 2019-02-05 | 中山大学 | A kind of image search method and system based on hand drawing style recommendation |
| CN109325529A (en) * | 2018-09-06 | 2019-02-12 | 安徽大学 | A kind of application of sketch recognition method and this method in commodity retrieval |
| CN109491327A (en) * | 2017-09-13 | 2019-03-19 | 发那科株式会社 | Threedimensional model producing device |
| CN109741454A (en) * | 2018-11-20 | 2019-05-10 | 浙江大学 | A kind of sketch formula three-dimensional cartoon model formative method based on component combination and deformation |
| CN110019914A (en) * | 2018-07-18 | 2019-07-16 | 王斌 | A kind of three-dimensional modeling data storehouse search method for supporting three-dimensional scenic interaction |
| CN110544300A (en) * | 2019-09-05 | 2019-12-06 | 佛山市玖章智能科技有限公司 | Method for automatically generating three-dimensional model based on two-dimensional hand-drawn image characteristics |
| CN110598634A (en) * | 2019-09-12 | 2019-12-20 | 山东文多网络科技有限公司 | Machine room sketch identification method and device based on graph example library |
| US10521700B2 (en) | 2017-12-14 | 2019-12-31 | Honda Motor Co., Ltd. | Methods and systems for converting a line drawing to a rendered image |
| CN110889854A (en) * | 2019-10-16 | 2020-03-17 | 深圳信息职业技术学院 | Sketch component segmentation method, system and device based on multi-scale deep learning and storage medium |
| CN112102284A (en) * | 2020-09-14 | 2020-12-18 | 推想医疗科技股份有限公司 | Marking method, training method and device of training sample of image segmentation model |
| CN112764538A (en) * | 2021-01-13 | 2021-05-07 | 杭州师范大学 | Gesture interaction based space capacity improving method in VR environment |
| CN112966138A (en) * | 2021-02-22 | 2021-06-15 | 济南大学 | Two-dimensional shape retrieval method and system based on contour feature point matching |
| CN113093904A (en) * | 2021-03-26 | 2021-07-09 | 河北建筑工程学院 | Method, system, medium and device for obtaining image based on somatosensory device |
| CN114913330A (en) * | 2022-07-18 | 2022-08-16 | 中科视语(北京)科技有限公司 | Point cloud component segmentation method and device, electronic equipment and storage medium |
| CN116597163A (en) * | 2023-05-18 | 2023-08-15 | 广东省旭晟半导体股份有限公司 | Infrared optical lens and method for manufacturing the same |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7729531B2 (en) * | 2006-09-19 | 2010-06-01 | Microsoft Corporation | Identifying repeated-structure elements in images |
| CN103942795A (en) * | 2014-04-22 | 2014-07-23 | 浙江大学 | Structural synthesis method of image object |
-
2015
- 2015-05-22 CN CN201510267873.3A patent/CN104850633B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7729531B2 (en) * | 2006-09-19 | 2010-06-01 | Microsoft Corporation | Identifying repeated-structure elements in images |
| CN103942795A (en) * | 2014-04-22 | 2014-07-23 | 浙江大学 | Structural synthesis method of image object |
Non-Patent Citations (2)
| Title |
|---|
| 万丽莉等: "一种基于部件空间分布的三维模型检索方法", 《软件学报》 * |
| 张必欢: "基于手绘草图的三维模型检索***", 《中国优秀硕士学位论文全文库》 * |
Cited By (35)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105243137B (en) * | 2015-09-30 | 2018-12-11 | 华南理工大学 | A kind of three-dimensional model search viewpoint selection method based on sketch |
| CN105243137A (en) * | 2015-09-30 | 2016-01-13 | 华南理工大学 | Draft-based three-dimensional model retrieval viewpoint selection method |
| CN106126581A (en) * | 2016-06-20 | 2016-11-16 | 复旦大学 | Cartographical sketching image search method based on degree of depth study |
| CN106126581B (en) * | 2016-06-20 | 2019-07-05 | 复旦大学 | Cartographical sketching image search method based on deep learning |
| CN106530397A (en) * | 2016-10-13 | 2017-03-22 | 成都希盟泰克科技发展有限公司 | Geological surface three-dimensional reconstruction method based on sparse profile geological contours |
| CN106530397B (en) * | 2016-10-13 | 2019-06-11 | 成都希盟泰克科技发展有限公司 | A kind of geology face three-dimensional rebuilding method based on sparse section geology contour line |
| CN106874350A (en) * | 2016-12-27 | 2017-06-20 | 合肥阿巴赛信息科技有限公司 | A kind of diamond ring search method and system based on sketch and distance field |
| CN107122396B (en) * | 2017-03-13 | 2019-10-29 | 西北大学 | Method for searching three-dimension model based on depth convolutional neural networks |
| CN107122396A (en) * | 2017-03-13 | 2017-09-01 | 西北大学 | Three-dimensional model searching algorithm based on depth convolutional neural networks |
| CN107220277A (en) * | 2017-04-14 | 2017-09-29 | 西北大学 | Image retrieval algorithm based on cartographical sketching |
| CN109491327B (en) * | 2017-09-13 | 2023-04-07 | 发那科株式会社 | Three-dimensional model making device |
| CN109491327A (en) * | 2017-09-13 | 2019-03-19 | 发那科株式会社 | Threedimensional model producing device |
| CN108109205B (en) * | 2017-12-08 | 2019-12-20 | 山东师范大学 | Incomplete model retrieval and reconstruction method based on hole filling |
| CN108109205A (en) * | 2017-12-08 | 2018-06-01 | 山东师范大学 | A kind of incomplete model index and method for reconstructing based on holes filling |
| US10521700B2 (en) | 2017-12-14 | 2019-12-31 | Honda Motor Co., Ltd. | Methods and systems for converting a line drawing to a rendered image |
| CN108170823A (en) * | 2018-01-04 | 2018-06-15 | 江西师范大学 | A kind of Freehandhand-drawing interactive three-dimensional model retrieval method understood based on high-level semantic attribute |
| CN110019914A (en) * | 2018-07-18 | 2019-07-16 | 王斌 | A kind of three-dimensional modeling data storehouse search method for supporting three-dimensional scenic interaction |
| CN109325529A (en) * | 2018-09-06 | 2019-02-12 | 安徽大学 | A kind of application of sketch recognition method and this method in commodity retrieval |
| CN109325529B (en) * | 2018-09-06 | 2021-07-30 | 安徽大学 | Sketch identification method and application of sketch identification method in commodity retrieval |
| CN109308324A (en) * | 2018-09-08 | 2019-02-05 | 中山大学 | A kind of image search method and system based on hand drawing style recommendation |
| CN109741454A (en) * | 2018-11-20 | 2019-05-10 | 浙江大学 | A kind of sketch formula three-dimensional cartoon model formative method based on component combination and deformation |
| CN110544300A (en) * | 2019-09-05 | 2019-12-06 | 佛山市玖章智能科技有限公司 | Method for automatically generating three-dimensional model based on two-dimensional hand-drawn image characteristics |
| CN110544300B (en) * | 2019-09-05 | 2021-06-29 | 佛山市玖章智能科技有限公司 | Method for automatically generating three-dimensional model based on two-dimensional hand-drawn image characteristics |
| CN110598634A (en) * | 2019-09-12 | 2019-12-20 | 山东文多网络科技有限公司 | Machine room sketch identification method and device based on graph example library |
| CN110598634B (en) * | 2019-09-12 | 2020-08-07 | 山东文多网络科技有限公司 | Machine room sketch identification method and device based on graph example library |
| CN110889854A (en) * | 2019-10-16 | 2020-03-17 | 深圳信息职业技术学院 | Sketch component segmentation method, system and device based on multi-scale deep learning and storage medium |
| CN110889854B (en) * | 2019-10-16 | 2023-12-05 | 深圳信息职业技术学院 | Sketch part segmentation method, system, device and storage medium based on multi-scale deep learning |
| CN112102284B (en) * | 2020-09-14 | 2024-05-28 | 推想医疗科技股份有限公司 | Marking method, training method and device for training sample of image segmentation model |
| CN112102284A (en) * | 2020-09-14 | 2020-12-18 | 推想医疗科技股份有限公司 | Marking method, training method and device of training sample of image segmentation model |
| CN112764538A (en) * | 2021-01-13 | 2021-05-07 | 杭州师范大学 | Gesture interaction based space capacity improving method in VR environment |
| CN112966138A (en) * | 2021-02-22 | 2021-06-15 | 济南大学 | Two-dimensional shape retrieval method and system based on contour feature point matching |
| CN112966138B (en) * | 2021-02-22 | 2022-04-12 | 济南大学 | Two-dimensional shape retrieval method and system based on contour feature point matching |
| CN113093904A (en) * | 2021-03-26 | 2021-07-09 | 河北建筑工程学院 | Method, system, medium and device for obtaining image based on somatosensory device |
| CN114913330A (en) * | 2022-07-18 | 2022-08-16 | 中科视语(北京)科技有限公司 | Point cloud component segmentation method and device, electronic equipment and storage medium |
| CN116597163A (en) * | 2023-05-18 | 2023-08-15 | 广东省旭晟半导体股份有限公司 | Infrared optical lens and method for manufacturing the same |
Also Published As
| Publication number | Publication date |
|---|---|
| CN104850633B (en) | 2018-10-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104850633A (en) | Three-dimensional model retrieval system and method based on parts division of hand-drawn draft | |
| US11586637B2 (en) | Search result processing method and apparatus, and storage medium | |
| WO2022116537A1 (en) | News recommendation method and apparatus, and electronic device and storage medium | |
| CN102521366B (en) | Image retrieval method integrating classification with hash partitioning and image retrieval system utilizing same | |
| CN107833213B (en) | Weak supervision object detection method based on false-true value self-adaptive method | |
| US9875253B2 (en) | Color sketch image searching | |
| CN106095893A (en) | A kind of cross-media retrieval method | |
| CN104850822B (en) | Leaf identification method under simple background based on multi-feature fusion | |
| CN109829065B (en) | Image retrieval method, device, equipment and computer readable storage medium | |
| CN107357765B (en) | Word document flaking method and device | |
| CN105740378B (en) | Digital pathology full-section image retrieval method | |
| CN110032650B (en) | Training sample data generation method and device and electronic equipment | |
| CN109857957B (en) | Method for establishing label library, electronic equipment and computer storage medium | |
| CN106844482B (en) | Search engine-based retrieval information matching method and device | |
| CN103455528A (en) | Handwritten character retrieval apparatus, handwritten character retrieval method and recording medium | |
| CN113239818B (en) | Table cross-modal information extraction method based on segmentation and graph convolution neural network | |
| CN102360436B (en) | Identification method for on-line handwritten Tibetan characters based on components | |
| CN110795613A (en) | Commodity searching method, device and system and electronic equipment | |
| Pengcheng et al. | Fast Chinese calligraphic character recognition with large-scale data | |
| CN116629258B (en) | Structured analysis method and system for judicial document based on complex information item data | |
| CN112613451A (en) | Modeling method of cross-modal text picture retrieval model | |
| CN104050461A (en) | Complex 3D motion recognition method and device | |
| CN105844299A (en) | Image classification method based on bag of words | |
| CN108241650B (en) | Training method and device for training classification standard | |
| Pancaroglu | Artist, style and year classification using face recognition and clustering with convolutional neural networks |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| EXSB | Decision made by sipo to initiate substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |