CN104392453B - The ransac characteristic matching optimization methods of image are inserted based on polar curve - Google Patents
The ransac characteristic matching optimization methods of image are inserted based on polar curve Download PDFInfo
- Publication number
- CN104392453B CN104392453B CN201410720152.9A CN201410720152A CN104392453B CN 104392453 B CN104392453 B CN 104392453B CN 201410720152 A CN201410720152 A CN 201410720152A CN 104392453 B CN104392453 B CN 104392453B
- Authority
- CN
- China
- Prior art keywords
- image
- polar curve
- matching
- ransac
- quality
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Expired - Fee Related
Links
- 238000000034 method Methods 0.000 title claims abstract description 44
- 238000005457 optimization Methods 0.000 title claims abstract description 12
- 239000011159 matrix material Substances 0.000 claims abstract description 25
- 238000012545 processing Methods 0.000 claims abstract description 8
- 238000012360 testing method Methods 0.000 claims abstract description 8
- 238000001514 detection method Methods 0.000 claims abstract description 5
- 238000005070 sampling Methods 0.000 claims description 8
- 230000006978 adaptation Effects 0.000 claims description 3
- HUTDUHSNJYTCAR-UHFFFAOYSA-N ancymidol Chemical compound C1=CC(OC)=CC=C1C(O)(C=1C=NC=NC=1)C1CC1 HUTDUHSNJYTCAR-UHFFFAOYSA-N 0.000 claims description 2
- 238000012216 screening Methods 0.000 abstract description 2
- 241000196324 Embryophyta Species 0.000 abstract 1
- 238000003780 insertion Methods 0.000 description 6
- 230000037431 insertion Effects 0.000 description 6
- 238000005516 engineering process Methods 0.000 description 4
- 230000008878 coupling Effects 0.000 description 2
- 238000010168 coupling process Methods 0.000 description 2
- 238000005859 coupling reaction Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000012549 training Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000000052 comparative effect Effects 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 230000000717 retained effect Effects 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/30—Determination of transform parameters for the alignment of images, i.e. image registration
Landscapes
- Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of ransac characteristic matching optimization methods that image is inserted based on polar curve, this method can improve quality of match, increase characteristic point high-quality number of matches, solve the problem of ransac algorithm image characteristic points high-quality is matched less, the degree of accuracy is not high.This method carries out feature point detection, description and matched to image to be matched first, carries out preliminary screening to matching set by ratio and symmetry test, weeds out erroneous matching;Secondly the basis matrix that original high-quality matching is gathered and it is supported is obtained using ransac algorithms;Recycle basis matrix to calculate the polar curve of characteristic matching point, and the polar curve of acquisition is selected, to ensure that polar curve is uniformly distributed on image as much as possible;Then the processing of thickness and polar curve quantity is carried out to the polar curve selected, is inserted into image;Finally carry out the acquisition process of high-quality matching based on ransac algorithms again to the image after processing.This method works well, it is adaptable to various images.
Description
Technical field
The present invention relates to computer image processing technology field, more particularly to a kind of ransac that image is inserted based on polar curve
Feature Points Matching optimization method.
Background technology
In computer vision, the concept of images match is widely used in object identification, vision tracking, three-dimensional reconstruction etc.
Problem.It depends on such idea, i.e., first with property detector (present invention uses surf property detectors) detection image
Upper some special points, then they are described, finally characteristic point is matched using the content of description, so as to realize figure
Picture and matching between image.
Images match can be roughly divided into the matching based on pixel grey scale and the matching based on characteristics of image, and wherein image is special
Levy and be divided into provincial characteristics, edge feature and point feature.The image matching technology of distinguished point based can reach it is effective, quick and
The basic demand of the high images match of robustness, therefore be widely used.At present, distinguished point based (ORB, BRISK, FAST,
SURF, SIFT etc.) image matching technology apply between different images, due to these images exist yardstick, the anglec of rotation,
Brightness, block and the change such as mirror-reflection, therefore after preliminary matches, there is substantial amounts of matching inferior.Matching result
It is inaccurate to cause the problems such as such as BREAK TRACK, image mosaic are lopsided, three-dimensional reconstruction effect is poor.Draw in the above-described techniques
Enter stochastical sampling unification algorism (i.e.:Ransac), characteristics of image can be more reliably matched, quality of match is improved.Its base
This thought is randomly to choose some matchings pair, and their corresponding relation is calculated according to the polarity constraint of double-visual angle.Afterwards, match
Remaining matching subset is all used for supporting such relation in set, final to obtain the matching set that maximum supports this relation.
Ransac algorithms have the characteristic for supporting that the more big possibility that is more accurate, obtaining correct result of set is bigger.But, this method is obtained
The number of matches arrived is on the low side, it is impossible to meet application demand.And the present invention the problem of can solve above well.
The content of the invention
Present invention aims at solve the problem of ransac algorithm image characteristic points high-quality is matched less, the degree of accuracy is not high.
Need in Matching supporting set containing this requirement of more multielement, according to the principle of ransac algorithms, carry for ransac algorithms
A kind of Feature Points Matching optimization method is gone out.This method increase based on surf characteristic points high-quality matching to quantity, be applicable
In various view data, such as:To large scene, part, noisy acoustic image etc., the result obtained by this method can be applied
In three-dimensional reconstruction, target following, in the technical field such as recognition of face.
The technical scheme adopted by the invention to solve the technical problem is that:A kind of ransac spies that image is inserted based on polar curve
Levy matching optimization method.This method obtains basis matrix first with ransac algorithms, on this basis feature based match point,
Accurate polar curve set is obtained by polarity geometrical relationship.Then suitable polar curve is selected from polar curve set.The mark that polar curve is selected
Standard is that the polar curve picked out can be uniformly distributed on two width figures to be matched as much as possible.The polar curve of selection is inserted into image
In, set the pixel of line region on image different from other parts.Surf characteristic points finally are re-started to image to be matched
Detection, description, arest neighbors matching, ratio testing, symmetrical test obtains more high-quality set of matches by ransac algorithms
Close.
In above process, the principle of surf characteristic matchings is carried out according to the Euclidean distance between surf Feature Descriptors
Expression;The principle of ransac algorithms is to select 8 matchings at random in multiple times (i.e.:Calculate the quantity for obtaining basis matrix), calculate
Their basis matrix, remaining set of matches is used for supporting this basis matrix in set, and finally retaining has maximum matching branch
The basis matrix of set is held, and returns to its Matching supporting set;The polarity geometrical principle of double-visual angle is the spy in present image
Levy a process space projection and straight line is formed in correspondence image, this straight line is exactly the polar curve of current signature point.
Method flow:
Step 1:Two images (image 1 and image 2) to be matched are read, the initial matching of two images to be matched is obtained
Set;
Step 1-1:Detect the characteristic point of two images respectively with surf property detectors;
Step 1-2:Calculate description of characteristic point in two images respectively with surf describers;
Step 1-3:Bi-directional matching is carried out to description using adaptation, each characteristic point of image 1 is found to image 2
Two best match, find two best match of each characteristic point in the image 1 in image 2;
Step 1-4:Ratio testing, handles two matching set (i.e. respectively:Image 1 arrives the matching set of image 2, and
Image 2 arrives the matching set of image 1), the distance ratio that Optimum Matching is matched with suboptimum is calculated, ratio is removed and is more than given threshold value
Matching;
Step 1-5:Symmetry is tested, when the index value in two matching set is symmetrical, extracts this matching set,
Asymmetric matching set is removed, symmetrical matching set is returned;
Step 2:Using ransac stochastical sampling unification algorisms, the maximum basis matrix for supporting matching to gather is calculated, is returned
The high-quality matching set for meeting this basis matrix and the basis matrix for supporting this characteristic matching collection;
Step 3:The basis matrix obtained using former algorithm, calculates polar curve of the match point in correspondence image;
Step 4:In polar curve set obtain can on image equally distributed polar curve;
Step 4-1:Limit is calculated, judges limit in image or outside image;
Step 4-2:When limit is outside image, according to polar curve and the intersection point relation of image border, 1,2,3 or 4 are chosen respectively
Bar polar curve makes a search;
Step 4-3:When limit is in image, according to the angled relationships between polar curve, choose 1 respectively, 2,3 or 4 polar curves
Make a search;
Step 5:Image is handled using the polar curve of selection;
Step 5-1:By 1,2,3 or 4 polar curves are inserted on image 1 and image 2 to be matched;
Step 5-2:Image pixel at polar curve is arranged to 0, and by the thickness of polar curve be respectively set to 1,5,10 or
The different-thickness of 15 pixels;
Step 6:The calculating of high-quality set of matches is re-started to the image after processing;
Step 6-1:The operating process of above-mentioned steps 1 is carried out, initial matching set is obtained;
Step 6-2:The operating process of above-mentioned steps 2 is carried out, high-quality matching set is obtained.
Beneficial effect:
1st, the present invention can obtain the matching set that more, quality is more excellent, the degree of accuracy is higher.
2nd, the present invention can be applied to any image.
3rd, the present invention is when selecting polar curve set, using simple step pick out can on image it is near uniform
The polar curve of distribution;No matter limit is that in image or outside image, polar curve whether there is, and whether polar curve is uniformly distributed, can be right
It carries out good screening;Different from former ransac algorithms, the present invention is directed to the polar curve of varying number, to different thickness at polar curve
Region is spent, can analyze and obtain high-quality matching set.
Brief description of the drawings
Fig. 1 is flow chart of the method for the present invention.
Wherein, dotted arrow represents to include the former algorithm flow of part;Solid arrow represents the flow that new algorithm is newly increased.
Fig. 2 is algorithm details flow chart involved in the present invention.
Fig. 3 is the details flow chart of the relevant polar curve processing of flow chart proposed by the present invention.
Embodiment
In order to be better understood from the optimization method that the present invention is realized, below in conjunction with the accompanying drawings, to the specific embodiment party of the present invention
Formula is further described.The language that the example of implementation is used in the de-scription does not cause limiting to the claimed invention.
Embodiment one
As shown in figure 1, the idiographic flow that the application present invention carries out characteristics of image Optimized Matching is as follows:
Step 1:Basis matrix is sought, the high-quality set of matches of former algorithm is obtained, details flow is as shown in Figure 2;
Step 1-1:Input two images image1, image2 to be matched;
Step 1-2:The key point keypoints of two images is detected with surf detectors respectively, surf describers are used
Calculate keypoints 64 dimension description;
Step 1-3:Bi-directional matching is carried out to description using adaptation.I.e.:The each characteristic point for finding image1 is arrived
Image2 two best match, then find two best match of each characteristic point in image1 in image2.
Step 1-4:Carry out ratio test.I.e.:Two matching set are handled respectively, to two present in each matching set
Individual matching value, calculates the distance ratio of two values, when ratio is more than rate threshold ratioTH (0.5f), deletes this matching, small
Retain in ratioTH matching;
Step 1-5:Carry out Symmetry Detection.I.e. two set will be met:(1) the current matching value of each matching set has
Two;The reference key value of (2) matching set training index value corresponding with another matching set is equal, while one
The training index value of individual matching set is equal with corresponding reference key value in another matching set.The matching that condition is all met
Retained, be used as matched well collection;
Step 1-6:Calculated using stochastical sampling unification algorism ransac.8 are repeatedly selected at random in matching set
With calculate basis matrix, once calculate basis matrix, in set it is remaining it is all matching all will polarity corresponding with matrix about
Shu Jinhang is tested, and the maximum support set that basis matrix is obtained returns to this basis i.e. as final high-quality set of matches
Matrix;
Step 2:The treatment of details flow of polar curve selection as shown in figure 3, calculate image characteristic point correspondence polar curve, selection respectively
The polar curve of varying number, these polar curves are required to be uniformly distributed on image as far as possible;
Step 2-1:Calculate polar curve.The basis matrix tried to achieve according to step 1 and Feature Points Matching set, with formula (1)
Match point of the match point on image1 on the polar curve polars0, image2 on image image2 is calculated in image image1
On polar curve polars1;
The expression equation such as following formula of polar curve:
Polars [n] [0] x+polars [n] [1] y+polars [n] [2]=0 (1)
Wherein y=image.rows, x=image.cols, n are polar curve quantity.
Step 2-2:Calculate the intersection point of polar curve limit, polar curve and edge;
With formula (2) calculate polar curve limit (x0, y0), wherein lines [j]=polars0 [0], lines [k]=
polars0][1]。
The intersection point of polar curve and image border x=0 is sought with formula (3).
The intersection point with edge y=0 is calculated with formula (4).
Step 2-3:If limit x0>0 and y0>0, then limit is in image, and otherwise limit is outside polar curve;
Step 2-4:When limit is outside image, judges polar curve quantity i (1,2,3,4), the polar curve of two images is done respectively
Following selection:During i=1, output line s_line [0] is the intersection point y in all polar curves with edgel(or xc) near 1/2 edge
Place;During i=2, output line s_line [0] is the intersection point y in all polar curves with edgel(or xc) in the pole near 1/3 edge
Line, s_line [1] is close to s_line [0] 2 times of distant places;During i=3, output line s_line [0] be all polar curves in edge
Intersection point yl(or xc) near 1/4 edge, s_line [1] is that s_line [2] is close to s_line [0] 2 times of distant places
Close to s_line [1] 3/2 times of distant place;During i=4, output line s_line [0] is the intersection point y in all polar curves with edgel(or
xc) near 1/5 edge, s_line [1] is that, close to s_line [0] 2 times of distant places, s_line [2] is close to s_line
[1] 3/2 times of distant place, s_line [3] is close to s_line [1] 4/3 times of distant place;
Step 2-5:When limit is in image, polar curve quantity i (1,2,3,4) is judged, according to the polar curve angle meter of formula (5)
Calculate formula and following selection is done to the polar curve of two images respectively:During i=1, output line s_line [0] be all polar curves in edge
Intersection point yl(or xc) near 1/2 edge;During i=2, output line s_line [0] is the intersection point in all polar curves with edge
yl(or xc) in the polar curve near 1/2 edge, s_line [1] is nearly 90 ° of the angle with s_line [0];During i=3, output
Line s_line [0] is the intersection point y in all polar curves with edgel(or xc) in the polar curve near 1/2 edge, s_line [1] is
Nearly 60 ° with s_line [0] angle, s_line [2] is nearly 120 ° of the angle with s_line [0];During i=4, output line s_line
[0] it is the intersection point y in all polar curves with edgel(or xc) in the polar curve near 1/2 edge, s_line [1] is and s_line
[0] nearly 45 ° of angle, s_line [2] is nearly 90 ° of the angle with s_line [0], and s_line [3] is the angle with s_line [2]
Nearly 135 °;
Step 3:Image is handled as follows with the polar curve of selection:
Step 3-1:Polar curve is inserted into image, 0 is set to for the pixel on image at polar curve, to different-thickness
Polar curve is operated.When polar curve thickness is 1 pixel, the pixel value on polar curve is 0;When polar curve thickness is 5 pixels, including pole
5 pixel values up and down of line are set to 0;When polar curve thickness is 10 pixels, including 10 pixel values up and down of polar curve are set to
0;
Step 4:The operation of step 1 is re-started to the image after processing, final high-quality matching set is obtained;
So far, the present invention completes the images match optimization method split based on polar curve, obtains more high-quality matchings
Set.
In order to it is clearer understand the present invention superiority, specific steps in conjunction with the embodiments, be listed below the present invention and
Existing former method is using one group of image log according to the comparative result in terms of Image Feature Point Matching.
Embodiment two
When fixed pole line number amount is 2, the present invention relatively includes with former method:
Result shown in table 1 represents to insert in the picture after 2 polar curves, when polar curve thickness be 1,5,10,15 or 20 pictures
When plain, (thickness is the 0 original calculation for representing no introducing polar curve to the characteristic point quantity and final high-quality number of matches detected
Method).
As shown in Table 1, without introducing in the case of the primal algorithm of polar curve, Fig. 1 and Fig. 2 feature points are respectively 1478
With 1452, coupling number is 15.When inserting polar curve, with the increase of thickness, characteristic point and coupling number are all presented first to increase and subtracted again
Small trend.When thickness is 10, high-quality matching can reach 22, and 47% is improved than primal algorithm.
Table 1:High-quality matching result compares when insertion polar curve thickness is different
Note:The quantity that data fix insertion polar curve is 2;Thickness is the polar curve thickness of insertion;Thicknes s=0
The matching result of former method is represented, that is, is not inserted into polar curve, directly entire image is operated.
Embodiment three
When fixed polar curve thickness is 5 pixels, the present invention relatively includes with former method:
Table 2 is the comparison of matching result under different situations when fixed polar curve thickness is 5 pixels.As can be seen that former method
(number=0) obtain high-quality matching be 15, and the present invention polar curve number be 1 to 4 between, when insertion polar curve number be
At 4, preferably, obtained high-quality matching can reach 23 to effect, improve 53%.
Table 2:High-quality matching result compares when inserting the quantity difference of polar curve
Note:The thickness that data fix insertion polar curve is 5 pixels;Number is the polar curve quantity of insertion;Nu mber=0
The matching result of former method is represented, that is, is not inserted into polar curve, directly entire image is operated.
From above-mentioned Tables 1 and 2:By the introducing of polar curve of the present invention, when the appropriate polar curve quantity of selection, and appropriate polar curve
During thickness, the result significantly improved than primal algorithm can be obtained.
Claims (4)
1. a kind of ransac characteristic matching optimization methods that image is inserted based on polar curve, it is characterised in that methods described is applied to
Three-dimensional reconstruction, target following, recognition of face, including step as described below:
Step 1:Two images to be matched are read, i.e.,:Image 1 and image 2, obtain the initial matching collection of two images to be matched
Close;
Step 2:Using ransac stochastical sampling unification algorisms, the maximum basis matrix for supporting matching to gather is calculated, returns and meets
The high-quality matching set of this basis matrix and the basis matrix for supporting matching set;
Step 3:The basis matrix obtained using ransac stochastical sampling unification algorisms, calculates pole of the match point in correspondence image
Line, is the acquisition process that polar curve set of the match point in correspondence image is calculated using basis matrix;
Step 4:In polar curve set obtain can on image equally distributed polar curve, suitable pole is selected in polar curve set
Line can be evenly distributed on image, and its process comprises the following steps:
Step 4-1:Limit is calculated, judges limit in image or outside image;
Step 4-2:When limit is outside image, according to polar curve and the intersection point relation of image border, choose 1 respectively, 2,3 or 4 poles
Line;
Step 5:Image is handled using the polar curve of selection, the polar curve selected is inserted in image, its polar curve it is treated
Journey following steps:
Step 5-1:By 1,2,3 or 4 polar curves are inserted on image to be matched;
Step 5-2:Image pixel at polar curve is arranged to 0, and the thickness of polar curve is respectively set to 1,5,10 or 15
The different-thickness of pixel;
Step 6:The calculating of high-quality matching set is re-started to the image after processing, high-quality is regained to treated image
The step of matching the process of set, its processing procedure is as follows:
Step 6-1:The operating process of above-mentioned steps 1 is carried out, initial matching set is obtained;
Step 6-2:The operating process of above-mentioned steps 2 is carried out, high-quality matching set is obtained.
2. a kind of ransac characteristic matching optimization methods that image is inserted based on polar curve according to claim 1, its feature
It is, the step 1 is the process that initial matching is obtained to image, is comprised the following steps:
Step 1-1:Detect the characteristic point of two images respectively with surf property detectors;
Step 1-2:Calculate description of characteristic point in two images respectively with surf describers;
Step 1-3:Bi-directional matching is carried out to description using adaptation, each characteristic point of image 1 is found to two of image 2
Best match, finds two best match of each characteristic point in the image 1 in image 2;
Step 1-4:Ratio testing, handles two matching set, i.e., respectively:Image 1 arrives the matching set of image 2, and image 2
To the matching set of image 1, the distance ratio that Optimum Matching is matched with suboptimum is calculated, that ratio is more than given threshold value is removed
Match somebody with somebody;
Step 1-5:Symmetry is tested, when the index value in two matching set is symmetrical, is extracted this matching set, is removed
Asymmetric matching set, returns to symmetrical matching set.
3. a kind of ransac characteristic matching optimization methods that image is inserted based on polar curve according to claim 1, its feature
It is, the step 2 is image characteristic point high-quality matching set and the basis matrix for obtaining ransac stochastical sampling unification algorisms
Process.
4. a kind of ransac characteristic matching optimization methods that image is inserted based on polar curve according to claim 1, its feature
It is, methods described obtains basis matrix first with ransac stochastical sampling unification algorisms, on this basis distinguished point based
Matching, accurate polar curve set is obtained by polarity geometrical relationship;Then suitable polar curve is selected from polar curve set, polar curve is selected
Standard be that the polar curve picked out can be uniformly distributed on two width figures to be matched as much as possible;The polar curve of selection is inserted into figure
As in, set the pixel of line region on image different from other parts;Surf features finally are re-started to image to be matched
Point detection, description, arest neighbors matching, ratio testing, symmetrical test is obtained more by ransac stochastical samplings unification algorism
High-quality matching set.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410720152.9A CN104392453B (en) | 2014-12-01 | 2014-12-01 | The ransac characteristic matching optimization methods of image are inserted based on polar curve |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410720152.9A CN104392453B (en) | 2014-12-01 | 2014-12-01 | The ransac characteristic matching optimization methods of image are inserted based on polar curve |
Publications (2)
Publication Number | Publication Date |
---|---|
CN104392453A CN104392453A (en) | 2015-03-04 |
CN104392453B true CN104392453B (en) | 2017-09-29 |
Family
ID=52610351
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201410720152.9A Expired - Fee Related CN104392453B (en) | 2014-12-01 | 2014-12-01 | The ransac characteristic matching optimization methods of image are inserted based on polar curve |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN104392453B (en) |
Families Citing this family (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106815824B (en) * | 2016-12-08 | 2019-07-05 | 华中科技大学 | A kind of image neighbour's optimization method improving extensive three-dimensional reconstruction efficiency |
CN113139986A (en) * | 2021-04-30 | 2021-07-20 | 东风越野车有限公司 | Integrated environment perception and multi-target tracking system |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US8401281B1 (en) * | 2003-10-20 | 2013-03-19 | Open Invention Network, Llc | Method and system for three dimensional feature attribution through synergy of rational polynomial coefficients and projective geometry |
CN104021542A (en) * | 2014-04-08 | 2014-09-03 | 苏州科技学院 | Optimization method of SIFT characteristic matching points based on limit restraint |
-
2014
- 2014-12-01 CN CN201410720152.9A patent/CN104392453B/en not_active Expired - Fee Related
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US8401281B1 (en) * | 2003-10-20 | 2013-03-19 | Open Invention Network, Llc | Method and system for three dimensional feature attribution through synergy of rational polynomial coefficients and projective geometry |
CN104021542A (en) * | 2014-04-08 | 2014-09-03 | 苏州科技学院 | Optimization method of SIFT characteristic matching points based on limit restraint |
Non-Patent Citations (2)
Title |
---|
"基于RANSAC算法的极线约束立体视觉匹配方法研究";张培耘等;《组合机床与自动化加工技术》;20131130(第11期);论文第20-22页及图5 * |
"基于双目视觉的图像匹配算法研究";李志;《中国优秀硕士学位论文全文数据库 信息科技辑》;20100615(第6期);论文第6页 * |
Also Published As
Publication number | Publication date |
---|---|
CN104392453A (en) | 2015-03-04 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
KR101165359B1 (en) | Apparatus and method for analyzing relation with image and image or video | |
JP5954712B2 (en) | Image processing apparatus, image processing method, and program thereof | |
JP5711387B2 (en) | Method and apparatus for comparing pictures | |
CN104504101B (en) | A kind of determination method and device of similar video | |
TWI469623B (en) | Method for distinguishing a 3d image from a 2d image and for identifying the presence of a 3d image format by feature correspondence determination | |
US10699156B2 (en) | Method and a device for image matching | |
US8983199B2 (en) | Apparatus and method for generating image feature data | |
TW201214335A (en) | Method and arrangement for multi-camera calibration | |
Lange et al. | Dld: A deep learning based line descriptor for line feature matching | |
TW201710989A (en) | System and method for determining whether a product image includes a logo pattern | |
Resch et al. | Scalable structure from motion for densely sampled videos | |
TW201143359A (en) | Method and apparatus for distinguishing a 3D image from a 2D image and for identifying the presence of a 3D image format by image difference determination | |
CN109766818A (en) | Pupil center's localization method and system, computer equipment and readable storage medium storing program for executing | |
CN104392453B (en) | The ransac characteristic matching optimization methods of image are inserted based on polar curve | |
CN108898597B (en) | Method for identifying painting and calligraphy based on smart phone | |
Cai et al. | An adaptive symmetry detection algorithm based on local features | |
CN113095385A (en) | Multimode image matching method based on global and local feature description | |
CN104850600B (en) | A kind of method and apparatus for searching for the picture comprising face | |
CN104866818B (en) | A kind of method and apparatus for searching for the picture comprising face | |
Liu et al. | Depth-aware layered edge for object proposal | |
Jorissen et al. | Multi-view wide baseline depth estimation robust to sparse input sampling | |
CN110599456B (en) | Method for extracting specific region of medical image | |
CN112991419A (en) | Parallax data generation method and device, computer equipment and storage medium | |
CN112396551A (en) | Watermark embedding method and device | |
CN107203755B (en) | Method, device and system for automatically adding new time sequence mark samples of remote sensing images |
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 | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20170929 |