CN106408022B - The sub- construction method of two valued description based on simple sample mode and three-valued strategy - Google Patents
The sub- construction method of two valued description based on simple sample mode and three-valued strategy Download PDFInfo
- Publication number
- CN106408022B CN106408022B CN201610832220.XA CN201610832220A CN106408022B CN 106408022 B CN106408022 B CN 106408022B CN 201610832220 A CN201610832220 A CN 201610832220A CN 106408022 B CN106408022 B CN 106408022B
- Authority
- CN
- China
- Prior art keywords
- point
- sampled
- groups
- valued
- characteristic point
- 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
- 238000010276 construction Methods 0.000 title claims abstract description 8
- 238000005070 sampling Methods 0.000 claims abstract description 19
- 238000000034 method Methods 0.000 claims abstract description 18
- 238000001514 detection method Methods 0.000 claims abstract description 5
- 238000009499 grossing Methods 0.000 claims abstract description 5
- 230000001174 ascending effect Effects 0.000 claims description 2
- 235000002918 Fraxinus excelsior Nutrition 0.000 claims 1
- 239000002956 ash Substances 0.000 claims 1
- 238000005516 engineering process Methods 0.000 description 3
- 230000010287 polarization Effects 0.000 description 3
- 230000003044 adaptive effect Effects 0.000 description 2
- 238000007667 floating Methods 0.000 description 2
- 101100126955 Arabidopsis thaliana KCS2 gene Proteins 0.000 description 1
- 241000132023 Bellis perennis Species 0.000 description 1
- 235000005633 Chrysanthemum balsamita Nutrition 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/74—Image or video pattern matching; Proximity measures in feature spaces
- G06V10/75—Organisation of the matching processes, e.g. simultaneous or sequential comparisons of image or video features; Coarse-fine approaches, e.g. multi-scale approaches; using context analysis; Selection of dictionaries
- G06V10/757—Matching configurations of points or features
Landscapes
- Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Theoretical Computer Science (AREA)
- Evolutionary Computation (AREA)
- Computing Systems (AREA)
- Databases & Information Systems (AREA)
- Artificial Intelligence (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The present invention is based on the sub- construction methods of two valued description of simple sample mode and three-valued strategy, it include: to acquire two images under Same Scene different perspectives and input computer, gray level image is converted by color image and carries out Gaussian smoothing, characteristic point is extracted respectively in two images using Harris Corner Detection Algorithm, calculate the principal direction of characteristic point, the sampled point in characteristic point sampling area is obtained to carry out smoothly to and to sampled point, 256 groups of sampled points pair are chosen from 400 groups of sampled point centerings, two valued description is constructed to for each characteristic point using 256 groups of sampled points, Feature Points Matching is carried out based on two valued description.Method provided by the invention calculates Image Feature Matching task simple, with preferable matching performance, suitable for mobile end equipment.
Description
Technical field
The present invention relates in image procossing in Feature Points Matching field, especially digital picture the building of two valued description and
Characteristic point matching method.
Background technique
Characteristic matching is the major issue of image procossing and computer vision field, and Matching Technology of Feature Point is known in target
Not, have in many scenes such as target tracking, scene splicing and be widely applied.Characteristics of image description and matched basic principle are:
It selects the regional area centered on characteristic point and characteristic matching is carried out according to texture information construction matching description in region.
The Matching Technology of Feature Point of mainstream is based on floating type description, and representative floating type matching description has SIFT [1], SURF
[2] and DAISY [3] etc..With the application of intelligent movable equipment, memory space is small, treatment effeciency is high two valued description at
For the technology being badly in need of at present.
Existing two valued description mainly has BRISK [4], FREAK [5] and BRIEF [6] etc..Wherein BRISK and
FREAK is sampled using fixed mode, is obtained the grayscale information at sampled point, is then compared simultaneously to sampled point gray value
Binaryzation comparison result, finally using the character string obtained after binaryzation as description.Both describe son be primarily present it is following
Problem: since sample template position is fixed, the grayscale information of specific position can only be obtained, can not be obtained according to local image characteristic
More useful information cause to describe sub- description power not high.BRIEF description determines the position of sampled point using stochastical sampling
It sets, but the point of directly stochastical sampling acquisition is more to redundancy, affects the matching performance of description.Meanwhile above-mentioned three kinds
Description carries out the two-value method that polarization is all employed when gray value a little pair compares --- and non-zero i.e. 1, this binarization method
It is highly unstable in the little flat site of gray value of image difference, cause the sub- performance of two valued description obtained also unstable.Cause
This, needs to study more effective, the more stable building of two valued description and feature matching method.
Bibliography:
1.D.Lowe,Distinctive image features from scale-invariant keypoints,
International Journal of Computer Vision.2004,60(2):91–110.
2.H.Bay,T.Tuytelaars and L.V.Gool,Speeded up robust features(SURF),
Computer Vision and Image Understanding.2008,110:346–359.
3.E.Tola,V.Lepetit and P.Fua,Daisy:An efficient dense descriptor
applied to wide-baseline stereo.IEEE Trans.on Pattern Analysis and Machine
Intelligence,2010,232(5):815–830.
4.S.Leutenegger,M.Chli,and R.Siegwart.Brisk:Binary robust invariant
scalable keypoints.International Conference on Computer Vision.2011,2548-
2555.
5.A.Ahi,R.Ortiz and P.Vandergheynst.FREAK:fast retina keypoint.IEEE
Conference on Computer Vision and Pattern Recognition.2012,2069-2076.
6.M.Calonder,V.Lepetit and M.Ozuysal,et al.BRIEF:Computing a local
binary descriptor very fast,IEEE Trans.on Pattern Analysis and Machine
Intelligence,2012,34(7):1281-1298
Summary of the invention
The present invention proposes that one kind is based on for the existing the disadvantages of sub- descriptive power of image two valued description is poor, performance is unstable
The sub- construction method of two valued description of simple sample mode and three-valued strategy, mainly comprises the steps that
Step S1: two images under Same Scene different perspectives are acquired and input computer；
Step S2: gray level image is converted by color image and carries out Gaussian smoothing；
Step S3: characteristic point is extracted respectively in two images using Harris Corner Detection Algorithm；
Step S4: the principal direction of characteristic point is calculated；
Step S5: obtaining the sampled point pair in characteristic point sampling area, and carries out to sampled point smooth；
Step S6: 256 groups of sampled points pair are chosen from 400 groups of sampled point centerings；
Step S7: two valued description is constructed to for each characteristic point using 256 groups of sampled points；
Step S8: Feature Points Matching is carried out based on two valued description.
Compared with current some methods using fixed sample mode, two valued description provided by the invention building side
Method, sample mode is simple, and sampled point is selected again after being randomly generated by Gaussian Profile, can be adaptive according to image content
The strong sampled point pair of the ability of portraying is selected on ground, and redundancy can be rejected under conditions of retaining effective information, improves description
The matching performance of son.Three-valued strategy is introduced when comparing result binaryzation, the binarization method for overcoming tradition polarization exists
Unstable disadvantage in image flat site.Compared to existing method, method provided by the invention is more accurate and stablizes.
Detailed description of the invention
Fig. 1 is that the present invention is based on the sub- construction method flow charts of the two valued description of simple sample mode and three-valued strategy.
Specific embodiment
As shown in Figure 1 for the present invention is based on the sub- construction method processes of the two valued description of simple sample mode and three-valued strategy
Figure mainly comprises the steps that two images under acquisition Same Scene different perspectives and inputs computer, converts color image
For gray level image and carries out Gaussian smoothing, extracts feature respectively in two images using Harris Corner Detection Algorithm
Point, obtain point to sampling configuration, the principal direction that calculates characteristic point, obtain sampled point in characteristic point sampling area to and to sampling
Point carry out it is smooth, chosen from 400 groups of sampled point centerings 256 groups of sampled points to, using 256 groups of sampled points to for each characteristic point structure
It builds two valued description, Feature Points Matching is carried out based on two valued description.The specific implementation details of each step are as follows:
Step S1: two images under Same Scene different perspectives are acquired and input computer.
Step S2: gray level image is converted by color image and carries out Gaussian smoothing.
Step S3: characteristic point is extracted respectively in two images using Harris Corner Detection Algorithm.
Step S4: calculating the principal direction of characteristic point, and concrete mode is, for any feature point F in two images, definition with
Centered on F, 23 for radius sampling area of the border circular areas as point F, be denoted as G (F), the gradient of calculating G (F) interior all pixels
Value obtains the gradient mean value [d of G (F)x,dy], by the corresponding direction θ=atan (d of the gradient mean valuey,dx) it is determined as characteristic point F
Principal direction.
Step S5: the sampled point pair in characteristic point sampling area is obtained, and sampled point is carried out smoothly, concrete mode is such as
Under:
Step S51: direction rotation into alignment is carried out to characteristic point sampling area, concrete mode is, in two images
The sampling area G (F) of point F is rotated clockwise the corresponding angle of F principal direction by any feature point F.
Step S52: obtaining the sampled point pair in characteristic point sampling area, and concrete mode is, in the sampling that step S51 is obtained
In region, 400 groups of sampled points pair for meeting Gaussian Profile are randomly generated.
Step S53: sampled point is carried out smoothly, concrete mode is that, for 800 sampled points of acquisition, will arrive point F distance
Less than 11 groups of samples at set be denoted as nearly center of circle point set, remaining groups of samples at set be denoted as remote center of circle point set；Make
Nearly centre point is carried out smoothly with the mean filter that radius is 1.5, the mean filter that actionradius is 2.5 is to remote centre point
It carries out smooth.
Step S6: 256 groups of sampled points pair are chosen from 400 groups of sampled point centerings, concrete mode is as follows:
Step S61: the comparison result of binaryzation sampled point pair, concrete mode are the 400 groups of points pair obtained for step S5
In any point to (pi,pj), compare sampled point piAnd pjGray value I (pi) and I (pj), if I (pi)>I(pj) then by the point
Pair comparison result be denoted as 1, be otherwise denoted as 0.
Step S62: the comparison result of storage sampled point pair, concrete mode is to create a table, each column pair in table
One group of sampled point pair is answered, totally 400 column；The value of each row represents the group point to the comparison result at different characteristic point under same column, is somebody's turn to do
The line number of table is equal to the number of characteristic point in two images.
Step S63: calculating variance and simultaneously select 256 groups of sampled points pair, concrete mode be the variance that is respectively arranged in computation sheet simultaneously
Non- ascending sort is carried out to each column according to variance size, picks out 256 groups of forward points pair of ranking results.
Step S7: two valued description is constructed to for each characteristic point using 256 groups of sampled points, concrete mode is, for appointing
One characteristic point, the 256 groups of sampled points obtained using step S6 are to two sampled points of more every group of sampled point centering as follows
Gray value, obtain one 3 dimension binary set:
The binary set of 256 groups of points pair is attached to obtain 768 dimension two-values of this feature point by wherein Δ value 10~15
Description.
Step S8: Feature Points Matching is carried out based on two valued description, concrete mode is, for any spy in the 1st width image
Levy point Fi, remember the 2nd width image in FiThe smallest characteristic point of Hamming distance is F between two valued descriptioni1, distance value is denoted as d1,
While and FiThe small characteristic point of Hamming distance time is F between two valued descriptioni2, distance value is denoted as d2If d1/d2Less than threshold
Value T, then by characteristic point (Fi,Fi1) be determined as one group of match point and export, wherein the value of T is 0.6~0.85.
Compared with current some methods using fixed sample mode, two valued description provided by the invention building side
Method, sample mode is simple, and sampled point is selected again after being randomly generated by Gaussian Profile, can be adaptive according to image content
The strong sampled point pair of the ability of portraying is selected on ground, and redundancy can be rejected under conditions of retaining effective information, improves description
The matching performance of son.Three-valued strategy is introduced when comparing result binaryzation, the binarization method for overcoming tradition polarization exists
Unstable disadvantage in image flat site.Compared to existing method, method provided by the invention is more accurate and stablizes.
Claims (1)
1. a kind of sub- construction method of two valued description based on simple sample mode and three-valued strategy, which is characterized in that this method
Specifically comprise the following steps:
Step S1: two images under Same Scene different perspectives are acquired and input computer；
Step S2: gray level image is converted by color image and carries out Gaussian smoothing；
Step S3: characteristic point is extracted respectively in two images using Harris Corner Detection Algorithm；
Step S4: calculating the principal direction of characteristic point, and concrete mode is that, for any feature point F in two images, definition is with F
Center, 23 are sampling area of the border circular areas of radius as point F, are denoted as G (F), calculate the gradient value of G (F) interior all pixels,
Obtain the gradient mean value [d of G (F)x,dy], by the corresponding direction θ=atan (d of the gradient mean valuey,dx) it is determined as characteristic point F's
Principal direction；
Step S5: the sampled point pair in characteristic point sampling area is obtained, and sampled point is carried out smoothly, concrete mode is as follows:
Step S51: direction rotation into alignment is carried out to characteristic point sampling area, concrete mode is, for any in two images
The sampling area G (F) of point F is rotated clockwise the corresponding angle of F principal direction by characteristic point F；
Step S52: obtaining the sampled point pair in characteristic point sampling area, and concrete mode is, in the sampling area that step S51 is obtained
In, 400 groups of sampled points pair for meeting Gaussian Profile are randomly generated；
Step S53: sampled point is carried out smoothly, concrete mode is that, for 800 sampled points of acquisition, will arrive point F distance and be less than
11 groups of samples at set be denoted as nearly center of circle point set, remaining groups of samples at set be denoted as remote center of circle point set；Use half
The mean filter that diameter is 1.5 carries out smoothly nearly centre point, and the mean filter that actionradius is 2.5 carries out remote centre point
Smoothly；
Step S6: 256 groups of sampled points pair are chosen from 400 groups of sampled point centerings, concrete mode is as follows:
Step S61: the comparison result of binaryzation sampled point pair, concrete mode are, for the step S5 400 groups of point centerings obtained
Any point is to (pi,pj), compare sampled point piAnd pjGray value I (pi) and I (pj), if I (pi)>I(pj) then by the point pair
Comparison result is denoted as 1, is otherwise denoted as 0；
Step S62: the comparison result of storage sampled point pair, concrete mode is to create a table, each column corresponding one in table
Group sampled point pair 400 arranges totally；The value of each row represents the group point to the comparison result at different characteristic point, the table under same column
Line number be equal to two images in characteristic point number；
Step S63: calculating variance and simultaneously select 256 groups of sampled points pair, concrete mode be the variance respectively arranged in computation sheet and according to
Variance size carries out non-ascending sort to each column, picks out 256 groups of forward points pair of ranking results；
Step S7: two valued description is constructed to for each characteristic point using 256 groups of sampled points, concrete mode is, for any spy
Point is levied, the 256 groups of sampled points obtained using step S6 are to the ashes of two sampled points of more every group of sampled point centering as follows
Angle value obtains one 3 dimension binary set:
The binary set of 256 groups of points pair is attached to obtain 768 dimension two valued descriptions of this feature point by wherein Δ value 10~15
Son；
Step S8: Feature Points Matching is carried out based on two valued description, concrete mode is, for any feature point in the 1st width image
Fi, remember the 2nd width image in FiThe smallest characteristic point of Hamming distance is F between two valued descriptioni1, distance value is denoted as d1, simultaneously
With FiThe small characteristic point of Hamming distance time is F between two valued descriptioni2, distance value is denoted as d2If d1/d2Less than threshold value T,
Then by characteristic point (Fi,Fi1) be determined as one group of match point and export, wherein the value of T is 0.6~0.85.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610832220.XA CN106408022B (en) | 2016-09-20 | 2016-09-20 | The sub- construction method of two valued description based on simple sample mode and three-valued strategy |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610832220.XA CN106408022B (en) | 2016-09-20 | 2016-09-20 | The sub- construction method of two valued description based on simple sample mode and three-valued strategy |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106408022A CN106408022A (en) | 2017-02-15 |
| CN106408022B true CN106408022B (en) | 2019-05-17 |
Family
ID=57996630
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610832220.XA Expired - Fee Related CN106408022B (en) | 2016-09-20 | 2016-09-20 | The sub- construction method of two valued description based on simple sample mode and three-valued strategy |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106408022B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111461285B (en) * | 2019-01-21 | 2024-03-05 | 京东科技控股股份有限公司 | Method and device for detecting electric equipment |
| CN114783002B (en) * | 2022-06-22 | 2022-09-13 | 中山大学深圳研究院 | Object intelligent matching method applied to scientific and technological service field |
| CN118396994B (en) * | 2024-06-26 | 2024-10-18 | 东莞市中钢模具有限公司 | Die-casting die adaptation degree detection method and system based on three-dimensional model |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101819636A (en) * | 2010-03-30 | 2010-09-01 | 河南理工大学 | Irregular area automatic matching method in the digital picture |
| CN104616300A (en) * | 2015-02-03 | 2015-05-13 | 清华大学 | Sampling mode separation based image matching method and device |
| CN104851094A (en) * | 2015-05-14 | 2015-08-19 | 西安电子科技大学 | Improved method of RGB-D-based SLAM algorithm |
| CN105740899A (en) * | 2016-01-29 | 2016-07-06 | 长安大学 | Machine vision image characteristic point detection and matching combination optimization method |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9158992B2 (en) * | 2013-03-14 | 2015-10-13 | Here Global B.V. | Acceleration of linear classifiers |
-
2016
- 2016-09-20 CN CN201610832220.XA patent/CN106408022B/en not_active Expired - Fee Related
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101819636A (en) * | 2010-03-30 | 2010-09-01 | 河南理工大学 | Irregular area automatic matching method in the digital picture |
| CN104616300A (en) * | 2015-02-03 | 2015-05-13 | 清华大学 | Sampling mode separation based image matching method and device |
| CN104851094A (en) * | 2015-05-14 | 2015-08-19 | 西安电子科技大学 | Improved method of RGB-D-based SLAM algorithm |
| CN105740899A (en) * | 2016-01-29 | 2016-07-06 | 长安大学 | Machine vision image characteristic point detection and matching combination optimization method |
Non-Patent Citations (2)
| Title |
|---|
| Binary Keypoint Descriptor for Accelerated Matching;Zhengguang Xu.etc;《2012 Fourth International Symposium on Information Science and Engineering》;20130411;第78-82页 * |
| 一种基于距离变换的不规则区域匹配算法;霍占强等;《计算机工程与科学》;20160731;第38卷(第7期);第1471-1478页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106408022A (en) | 2017-02-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104850850B (en) | A kind of binocular stereo vision image characteristic extracting method of combination shape and color | |
| CN110032983B (en) | Track identification method based on ORB feature extraction and FLANN rapid matching | |
| CN106408022B (en) | The sub- construction method of two valued description based on simple sample mode and three-valued strategy | |
| CN105654421B (en) | Based on the projective transformation image matching method for converting constant low-rank texture | |
| Lange et al. | Dld: A deep learning based line descriptor for line feature matching | |
| CN103400384A (en) | Large viewing angle image matching method capable of combining region matching and point matching | |
| CN106650580B (en) | Goods shelf quick counting method based on image processing | |
| CN111126412A (en) | Image key point detection method based on characteristic pyramid network | |
| CN110569861A (en) | Image matching positioning method based on point feature and contour feature fusion | |
| CN107862319B (en) | Heterogeneous high-light optical image matching error eliminating method based on neighborhood voting | |
| CN113095385B (en) | Multimode image matching method based on global and local feature description | |
| CN112614167A (en) | Rock slice image alignment method combining single-polarization and orthogonal-polarization images | |
| CN109086350B (en) | Mixed image retrieval method based on WiFi | |
| CN106408023B (en) | Image characteristic point two valued description and matching process based on group comparison strategy | |
| CN113554036A (en) | Characteristic point extraction and matching method for improving ORB algorithm | |
| CN110246165B (en) | Method and system for improving registration speed of visible light image and SAR image | |
| CN114066954B (en) | Feature extraction and registration method for multi-modal image | |
| Huang et al. | FAST and FLANN for feature matching based on SURF | |
| CN104036494A (en) | Fast matching computation method used for fruit picture | |
| CN108492256B (en) | Unmanned aerial vehicle video fast splicing method | |
| Liu et al. | Extend point descriptors for line, curve and region matching | |
| Chen et al. | Geometric and non-linear radiometric distortion robust multimodal image matching via exploiting deep feature maps | |
| CN107330436A (en) | A kind of panoramic picture SIFT optimization methods based on dimensional criteria | |
| CN106327423B (en) | Remote sensing image registration method and system based on directed line segment | |
| Wang et al. | Deep homography estimation based on attention mechanism |
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 | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20190517 |