CN105678733A - Infrared and visible-light different-source image matching method based on context of line segments - Google Patents
Infrared and visible-light different-source image matching method based on context of line segments Download PDFInfo
- Publication number
- CN105678733A CN105678733A CN201410669821.4A CN201410669821A CN105678733A CN 105678733 A CN105678733 A CN 105678733A CN 201410669821 A CN201410669821 A CN 201410669821A CN 105678733 A CN105678733 A CN 105678733A
- Authority
- CN
- China
- Prior art keywords
- line segment
- straight
- point
- infrared
- matching
- 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.)
- Pending
Links
Landscapes
- Image Analysis (AREA)
Abstract
The invention relates to an infrared and visible-light different-source image matching method based on the context of line segments, comprising the following steps: using an LSD algorithm to detect line segments in an image, and selecting key line segments according to geometric constraint rules; detecting corners through an improved image corner detection method; calculating the Manhattan distance between the line segments in four-quadrant neighborhoods of a feature point to get the contribution of each line segment to the feature point, and on the basis, building a feature descriptor based on the context of line segments through adoption of a circular array; and using a bidirectional matching strategy and an RANSAC algorithm to realize infrared and visible-light image matching. Through the method, more correct matching point pairs can be acquired. The method can adapt to exact matching of infrared and visible-light images with serious gray level difference, and is superior to mainstream different-source image matching algorithms in terms of robustness and time efficiency.
Description
Technical field
The present invention relates to technical field that is infrared and visible ray allos images match, specifically a kind of contextual infrared with visible ray allos image matching method based on straight line.
Background technology
Infrared mating with visible images is the important branch of allos images match, has important application in fields such as image co-registration, automatic target detection, change detections. Infrared sensor has the advantages such as all weather operations, immunity from interference are strong, and the image that visible light sensor obtains has features such as contrast gradient height, texture information are abundant, clear picture. In image co-registration field, in order to obtain more abundant scene information, it is necessary to realize information fusion on the infrared basis mated with visible images, strengthen complementary, reduce the uncertainty of scene analysis and understanding; In automatic target detection field, it is necessary to according to target information known in visible images, the method for images match is adopted to find out corresponding target from real-time infrared image. Due to the difference of the factors such as imaging equipment, spectrum used and shooting time, usually present complicated gray difference between infrared and visible images, bring bigger difficulty to coupling. Infrared is still a problem having challenge with mating of visible images.
Allos image matching method can be divided into the big class of the method two of the method based on region and feature based substantially. Method based on region adopts the similarity between the Similar measure definition matching areas such as gray scale difference, cross-correlation, mutual information to determine corresponding relation. But owing to usual grey scale change between infrared and visible images is complicated, gray-scale value is difficult to reflect the similarity of matching area. Although mutual information measure can in the grey scale change to a certain degree adapted between allos image, but this kind of method needs to adopt complicated searching algorithm, and efficiency is low, and the initialize required, otherwise easily it is absorbed in locally extreme value. The method of feature based is by extracting the stability features such as angle point, tapping point, difference of Gaussian extreme point, and construction feature descriptor realizes images match.Compared with the method based on region, the method for feature based has certain advantage in counting yield, deformation adaptive faculty and anti-partial occlusion etc., is usually used in allos images match. Traditional feature description operators such as () such as classical SIFT, SURF designs for homology images match, usually utilize the Gradient distribution attribute near unique point to build descriptor, not there is mode unchangeability, when for infrared mate with visible images time, often error hiding rate is higher, and even it fails to match. Scene boundaries generally corresponds to the edge in image (available straight-line segment approximate expression), and be present in allos image comparatively stablely.
Summary of the invention
For the deficiencies in the prior art, the present invention provides the matching process of a kind of infrared and visible images accurately and fast.
The technical scheme that the present invention adopts for achieving the above object is: a kind of contextual infrared with visible ray allos image matching method based on straight-line segment, comprises following process:
Step 1: adopt LSD algorithm to detect the straight-line segment published picture in picture, and pick out crucial straight-line segment according to geometrical constraint rule;
Step 2: by the image angular-point detection method detection angle point improved;
Step 3: by calculating the Manhattan distance of line segment in unique point four quadrant neighborhood, obtain every bar straight-line segment to the contribution of unique point, adopt the mode of circular array on this basis, builds based on contextual feature description of straight-line segment;
Step 4: use bi-directional matching strategy and RANSAC algorithm to realize infrared mating with visible images.
The true Harris angular-point detection method of described improvement is: be divided into by image with partly overlapping subregion, then in overall situation region, feature is not put significantly and is become significant unique point in local neighborhood, Harris angular-point detection method extract minutiae traditionally in above-mentioned every sub regions again, the unique point ensured in global scope is evenly distributed.
The described virtual angular-point detection method based on edge line is: picking out crucial straight-line segment in the edge line section detected according to geometrical constraint rule, extended along himself direction by these straight-line segments, the intersection point obtained is virtual angle point.
Described geometrical constraint rule comprises: the length of straight-line segment is not less than limit value lth; Angle between straight-line segment meets θth1< θ < θth2; Distance d between straight-line segment is no more than limit value dth。
The image angular-point detection method of described improvement comprise improvement true Harris angular-point detection method and based on the virtual angular-point detection method of edge line.
Described structure draws together following process based on line segment contextual feature description attached bag: distribute for single line segment, a straight-line segment proper vector abstract in space is represented, and based on the direction θ of the length l of line segment, line segment, line segment to distance d tri-attribute constructions scoring function of central point; For race's line segment distribution, every bar line segment decomposing corresponding quadrant, the line segment of the different quadrant of statistics distributes and calculates score respectively, distinguishes the spatial relation of different line segment with this.
Described straight-line segment context refers to: the line segment distribution in the local neighborhood centered by unique point, mainly comprise the distribution of single line segment and the distribution of race's line segment, wherein, the description of single line segment distribution, refers to and a straight-line segment proper vector abstract in space is represented; And race line segment describes, it it is the mathematical description in order to the position relation obtaining in space between different line segment.
Described bi-directional matching strategy is: performing just after coupling, exchange the role of image subject to registration and reference picture, identical unique point is re-executed matching algorithm, obtain negative relational matching point set.
The present invention has following useful effect and advantage:
The present invention, on the basis of feature point extraction, utilizes line segment context in unique point neighborhood to carry out construction feature descriptor, and adopts bi-directional matching strategy and RANSAC algorithm to realize infrared mating with visible images. Compared with the method for other feature based, the algorithm of the present invention can get how correct matching double points, the accurate of infrared and visible images that can adapt to gray difference comparatively serious mates, and is all better than main flow allos image matching algorithm in robustness and time efficiency.
Accompanying drawing explanation
Fig. 1 is the overall flow figure of the present invention;
Fig. 2 is LSD line segment detection result figure, and wherein (a) is infrared image, and (b) is visible images;
Fig. 3 is different Angular Point Extracting Method comparing result figure, and wherein (a) is Harris Corner Detection result, and (b) is the Harris Corner Detection result improved;
Fig. 4 is virtual angle point grid exemplary plot, and wherein (a) is infrared image, and (b) is visible images;
Fig. 5 is comprehensive Corner Detection result figure, and wherein (a) is infrared image, and (b) is visible images;
Fig. 6 is single line segment distribution plan;
Fig. 7 is line segment scoring function schema;
The line segment that Fig. 8 is different distributes for identical description figure;
Fig. 9 is four quadrant feature description graph;
Figure 10 (a) is Concentric circle array figure, and (b) adapts to rotate the Concentric circle array with Scale invariant;
Figure 11 is infrared with visible images matching result figure;
Figure 12 is first group of comparing result figure, and wherein (a) is LSS method, and (b) is LS method, and (c) is the inventive method;
Figure 13 is the 2nd group of comparing result figure, and wherein (a) is LSS method, and (b) is the inventive method for LS method, (c);
Figure 14 is the 3rd group of comparing result figure, and wherein (a) is LSS method, and (b) is the inventive method for LS method, (c).
Embodiment
Below in conjunction with drawings and Examples, the present invention is described in further detail.
The present invention in conjunction with unique point and linear edge, construction feature descriptor; Adopt Concentric circle array method, the sub-dimension of extended description, it is to increase the robustness of coupling; Finally, nearest neighbour matching process and RANSAC algorithmic match is utilized, it is achieved the benchmark figure of different modalities and mating of target figure, concrete treatment scheme is as shown in Figure 1.
1. line segment extraction
The present invention adopts LSD algorithm detection of straight lines section. The method robustness is relatively strong, and the straight-line segment of detection can well depend on true edge, and algorithm real-time is better.
Edge-description is become the region being made up of the pixel that gradient direction is identical by LSD, by region growing, the pixel of same gradient direction is polymerized to region, an edge, extracts straight-line segment from region. Fig. 2 is the infrared and visible ray linear edge adopting LSD algorithm to detect.
2. feature point detection
In order to improve the robustness of feature point detection, the present invention proposes the image angular-point detection method of a kind of improvement.
The 2.1 true Corner Detection improved
Traditional Harris angular-point detection method, by calculating the average gradient square matrices of each pixel, is analyzed its eigenwert and is obtained unique point. But for infrared image, the unique point that the method detects often distributes uneven. Therefore, the present invention proposes the Harris angular-point detection method of improvement.Image is divided into n partly overlapping subregion, Harris angular-point detection method extract minutiae is used in every sub regions, can ensure that in the overall situation feature is not put significantly in local neighborhood, to become significant unique point be detected, the unique point in global scope is evenly distributed. Adopt partly overlapping method, it is possible to effectively avoid the generation of the unique point missing inspection on zone iimit. Finally the unique point in smooth region is rejected, obtained stable unique point. As shown in Figure 3, wherein, Fig. 3 (a) is the detected result adopting Harris angle point method to obtain, Fig. 3 (b) is the result adopting the Harris angular-point detection method improved to obtain, it may be seen that adopt the Harris angular-point detection method improved can strengthen the reasonableness of angle point distribution.
2.2 virtual Corner Detection
Owing to the infrared true angle point distributional difference detected with visible images is relatively big, and linear edge distribution is relatively more consistent, and therefore, the present invention proposes a kind of virtual angular-point detection method based on edge line. Virtual angle point is the angle point not existed in image, is the imaginary intersection point obtained after the edge line section detected being extended, and they have same characteristic with angle point, are features more stable in image. The definition of virtual angle point must meet three constraints:
(1) length of straight-line segment is not less than limit value lth. Because long straight-line segment generally can reflect that the key structural feature of image and its extraction results contrast are sane.
(2) angle between straight-line segment meets θth1< θ < θth2. Due to the impact of image noise, the straight-line segment of extraction may on position or direction and real situation deviation to some extent, if close to parallel between straight-line segment, even if these deviations are very little, the intersection point of straight-line segment also can differ greatly with real situation.
(3) distance d between straight-line segment (two on straight-line segment distance) between most near point is no more than limit value dth. The restriction of the spacing of straight-line segment can retrain the straight-line segment logarithm of same bar straight-line segment composition, thus limits the virtual angle point number of extraction. Wherein, lth, θth1, θth2And dthIt is empirical constant Deng limit value, artificially sets according to feature of image. Fig. 4 (a) and Fig. 4 (b) blue markings point are the virtual angle point that simple infrared image and visible images extract respectively, it can be seen that both virtual angle points have higher consistence.
Finally, in conjunction with true angle point and virtual angle point, the visible ray that the present invention extracts and infrared image Corner Detection result are as shown in Figure 5. Fig. 5 (a) is infrared image, Fig. 5 (b) is visible images, white marking point represents the true angle point using the Harris angular-point detection method detection improved, blue markings point represents the virtual angle point of detection, it may be seen that the distribution of blue markings point has higher consistence.
3. feature description
The description of 3.1 single line segment distributions
As shown in Figure 6, to the distribution of single line segment, the present invention adopts the direction θ of the length l of line segment, line segment, line segment to distance d tri-attributes of central point, and based on these three attribute constructions a scoring function, shown in (1),
Wherein, d is the Manhattan distance AB+BC of an A to line segment CD, l is the length of line segment CD. In this research, why select Manhattan distance AB+BC, instead of Euclidean distance AB, it is because when the length of line segment red in Fig. 6 and line segment CD is equal, the score of the two is also equal. Therefore, Manhattan distance can effectively distinguish these two kinds different line segment distributions.
In addition, in order to describe the distribution of single line segment for unique point significantly, two fundamental principles are followed in the design of this scoring function:
1) line segment is more long, and score is more high;
2) line segment is more short to the distance of point of interest, and score is more high.
As can be seen from Figure 7, when l increases, f (d+l) moves downwards, and score increases; When d reduces, f (d) and f (d+l) moves up simultaneously, but f (d) is faster than the speed of f (d+l) movement, and score increases. Therefore, this scoring function meets this two principle of design, it is possible to the effective spatial distribution characterizing single straight line. Line segment score is decomposed x and y direction by the last direction according to line segment, characterizes the directional information of line segment.
The description of 3.2 races line segment distribution
Often more than one, the line segment in local neighborhood centered by unique point, in order to the position relation characterized in spatial neighborhood between different line segment, we need race's line segment distribution to be described, thus build contextual based on line segment, it is possible to characterize feature description of region lineal layout feature. As shown in Figure 8, when the length of two line segments, direction and to the distance of point of interest all identical time, their respective single line segment distribution scores are identical, identical score that the line segment distribution being also exactly different is corresponding. In order to distinguish the distribution of these two kinds of different line segments, the present invention proposes the method for four quadrant feature descriptions, as shown in Figure 9, every bar line segment is decomposed corresponding quadrant, the line segment distribution of the different quadrant of statistics respectively, distinguishes the spatial relation (in Fig. 9 spatial distribution can effectively distinguish) represented by red straight line of different line segment with this; Then, the score of all line segments of each quadrant is added, obtains the line segment distribution situation of each quadrant; Finally, this score is converted into feature description, in order to represent the line segment distribution situation in this region.
The feature description of 3.3 Shandong rods
The sub-structure method of traditional feature description centered by unique point, can only describe the line segment distribution situation in neighborhood near this unique point, and quantity of information is less. In order to improve the robustness of descriptor, the present invention proposes the method for Concentric circle array. Use for reference the thought constructing rectangular array subregion in SIFT method, as shown in Figure 10, we propose centered by unique point, concentric(al) circles is built with different radiuses, concentric(al) circles is chosen sampling point uniformly, to each sampling point, the method identical with unique point is adopted to be described, together with sampling point is combined in unique point, obtain a higher dimension descriptor, this descriptor also can react the distribution situation of all line segments around sampling point by response feature point, has higher robustness.
Compared with rectangular array in SIFT method, this circular array has stronger advantage, is mainly reflected in, it is possible to by the radius of the angle of adjustment point of interest and concentric(al) circles, adapt to the invariant feature of angle and yardstick.
Finally, above-mentioned feature description is normalized by we, to resist the impact of illumination variation.
4. characteristic matching
After the feature description subvector obtaining image, mate to the proper vector of two width images according to certain similarity measurement, with the one-to-one relationship obtained between image.
The present invention uses card side's distance as the standard of similarity measurement, and the formula of card side's distance is such as formula shown in (2)
Wherein, g (), h () refer to the infrared descriptor corresponding with visible images vector respectively, then use nearest neighbour time nearest neighbor distance to carry out characteristic matching than method. After the proper vector obtaining benchmark image and target image, get certain key point in benchmark image, find out the first two key point that card side in itself and target image is nearest.In these two key points, if nearest distance is less than certain threshold value with secondary near ratio, then accepting this to matching point, formula (3) is the mathematical notation of this kind of matching algorithm.
Wherein, DABAnd DACReferring to nearest neighbour and time nearest neighbor distance respectively, threshold is the empirical constant according to priori setting. But generally not there is larger difference in the unique point quantity of the image zooming-out of homology, the matching point that unidirectional coupling may only find part correct. Therefore, benchmark image and target image are carried out bi-directional matching by inventive algorithm. Here object is not the matching point rejecting mistake, but makes up in unidirectional coupling the matching point that may miss, to strengthen the robustness of algorithm. Performing just after coupling, exchange the role of image subject to registration and reference picture, identical unique point is re-executed above-mentioned algorithm, obtain negative relational matching point set. Finally, RANSAC algorithm is used in just reverse matching point set, chooses the consistent subset of maximum space, obtain total matching point set.
5. embodiment
Embodiment 1. matching result
In order to prove the validity of inventive algorithm, the hardware environment of emulation experiment example is CPUDual-Core2.5GHz, internal memory 2GB, WindowsXP+SP3, VisualStudio2013, and the image resolution rate that experimental example uses is 256 × 512. In experimental example, the number getting concentric(al) circles is 2, and on concentric(al) circles, the number of uniform sampling point is 8. When concentric(al) circles radius interval is less, sampling point, than comparatively dense, causes the degree of overlapping between sampling point local neighborhood excessive, and the repeatability of the different dimensions eigenwert of feature description is too high, makes the separating capacity of feature description more weak; When concentric(al) circles radius is bigger, sampling point is more sparse, the significant feature of part may be omitted, feature description is caused not possess the ability describing unique point neighborhood characteristics completely, making the consistence of feature description of different images correspondence position become weak, considering the radius arranging concentric(al) circles, to be spaced apart 25 be a rational value of comparison. Limit by length, only listed in experimental example comparatively typically infrared with visible images matching result. As shown in figure 11, it is the matching result using the inventive method to obtain.
Embodiment 2. comparing result
The method proposed in the present invention is compared with two kinds of classical multi-mode image matching algorithms based on descriptor, it is image matching method (the LocalSelfSimilarity based on local self similarity respectively, LSS) with based on wide baseline image matching process (LineSignature, LS) of line features. And from time complexity, robustness two aspects, algorithm performance is analyzed.
Each method matching result with visible images infrared to three groups is as shown in Figure 12～Figure 14. In first group of Matching Experiment, as shown in Figure 12 (b), owing to linear feature is not obvious or short lines section is more, LS method matching result is poor. In the 2nd group of Matching Experiment, as shown in Figure 13 (a), for the region that some structural informations are similar, LSS method matching result is undesirable. In the 3rd group of Matching Experiment, as shown in Figure 14 (b), owing to lacking the constraint of the overall situation, there is relatively large deviation in part line segment matching result. Under contrast, the method that the present invention proposes all achieves good matching precision.
In a word, LSS can not identify the region that some structural informations are similar, it is very difficult to realizes rotating the unchangeability with yardstick; LS has rotation and scale invariability, but not obvious for figure cathetus feature, or easily loses efficacy when short lines section is more, and depends on endpoint location, and the inaccurate meeting of endpoint location brings error; The inventive method goes for general situation, has higher robustness, and easily realizes rotating the unchangeability with yardstick. In addition, in time performance, the method for the present invention is also better than other two kinds of methods.
Claims (7)
1. one kind contextual infrared with visible ray allos image matching method based on straight-line segment, it is characterised in that: comprise following process:
Step 1: adopt LSD algorithm to detect the straight-line segment published picture in picture, and pick out crucial straight-line segment according to geometrical constraint rule;
Step 2: by the image angular-point detection method detection angle point improved;
Step 3: by calculating the Manhattan distance of line segment in unique point four quadrant neighborhood, obtain every bar straight-line segment to the contribution of unique point, adopt the mode of circular array on this basis, builds based on contextual feature description of straight-line segment;
Step 4: use bi-directional matching strategy and RANSAC algorithm to realize infrared mating with visible images.
2. according to claim 1 contextual infrared with visible ray allos image matching method based on straight-line segment, it is characterised in that: the image angular-point detection method of described improvement comprise improvement true Harris angular-point detection method and based on the virtual angular-point detection method of edge line.
3. according to claim 2 contextual infrared with visible ray allos image matching method based on straight-line segment, it is characterized in that, the true Harris angular-point detection method of described improvement is: be divided into by image with partly overlapping subregion, then in overall situation region, feature is not put significantly and is become significant unique point in local neighborhood, according to Harris angular-point detection method extract minutiae in every sub regions.
4. according to claim 2 contextual infrared with visible ray allos image matching method based on straight-line segment, it is characterized in that, the described virtual angular-point detection method based on edge line is: pick out crucial straight-line segment in the edge line section detected according to geometrical constraint rule, being extended along himself direction by key straight-line segment, the intersection point obtained is virtual angle point.
5. contextual infrared with visible ray allos image matching method based on straight-line segment according to claim 1 or 4, it is characterised in that, described geometrical constraint rule comprises: the length of straight-line segment is not less than length limit value lth; Angle between straight-line segment meets θth1< θ < θth2; Distance d between straight-line segment is no more than distance limit value dth。
6. according to claim 1 contextual infrared with visible ray allos image matching method based on straight-line segment, it is characterized in that, described structure draws together following process based on line segment contextual feature description attached bag: distribute for single line segment, a straight-line segment proper vector abstract in space is represented, and based on distance structure scoring function to central point of the length of straight-line segment, direction and straight-line segment; For race's line segment distribution, every bar straight-line segment decomposing corresponding quadrant, the straight-line segment of the different quadrant of statistics distributes and calculates score respectively, to distinguish the spatial relation of different straight-line segment.
7. according to claim 1 contextual infrared with visible ray allos image matching method based on straight-line segment, it is characterized in that: described bi-directional matching strategy is: performing just after coupling, exchange the role of image subject to registration and reference picture, identical unique point is re-executed matching algorithm, obtains negative relational matching point set.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410669821.4A CN105678733A (en) | 2014-11-21 | 2014-11-21 | Infrared and visible-light different-source image matching method based on context of line segments |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410669821.4A CN105678733A (en) | 2014-11-21 | 2014-11-21 | Infrared and visible-light different-source image matching method based on context of line segments |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN105678733A true CN105678733A (en) | 2016-06-15 |
Family
ID=56958361
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410669821.4A Pending CN105678733A (en) | 2014-11-21 | 2014-11-21 | Infrared and visible-light different-source image matching method based on context of line segments |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105678733A (en) |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107464257A (en) * | 2017-05-04 | 2017-12-12 | 中国人民解放军军械工程学院 | Wide baseline matching process and device |
| CN108205645A (en) * | 2016-12-20 | 2018-06-26 | 中国科学院沈阳自动化研究所 | A kind of reference map quality evaluating method of heterologous image matching system |
| CN110189284A (en) * | 2019-05-24 | 2019-08-30 | 南昌航空大学 | A kind of infrared and visible light image fusion method |
| CN110992407A (en) * | 2019-11-07 | 2020-04-10 | 武汉多谱多勒科技有限公司 | Infrared and visible light image matching method |
| CN111223134A (en) * | 2020-01-19 | 2020-06-02 | 北京航空航天大学 | Feature vector construction method of linear features, feature matching method and computing equipment |
| CN111354018A (en) * | 2020-03-06 | 2020-06-30 | 合肥维尔慧渤科技有限公司 | Object identification method, device and system based on image |
| CN112487233A (en) * | 2020-11-27 | 2021-03-12 | 重庆邮电大学 | Infrared and visible light image retrieval method based on feature decoupling |
| CN113095384A (en) * | 2021-03-31 | 2021-07-09 | 安徽工业大学 | Remote sensing image matching method based on context characteristics of straight line segments |
| CN113095385A (en) * | 2021-03-31 | 2021-07-09 | 安徽工业大学 | Multimode image matching method based on global and local feature description |
| CN113221923A (en) * | 2021-05-31 | 2021-08-06 | 西安电子科技大学 | Feature decomposition method and system for multi-mode image block matching |
| CN113792788A (en) * | 2021-09-14 | 2021-12-14 | 安徽工业大学 | Infrared and visible light image matching method based on multi-feature similarity fusion |
| CN116758067A (en) * | 2023-08-16 | 2023-09-15 | 梁山县成浩型钢有限公司 | Metal structural member detection method based on feature matching |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102663725A (en) * | 2012-03-05 | 2012-09-12 | 西北工业大学 | A visible light and SAR image registration method based on linear features and control points |
-
2014
- 2014-11-21 CN CN201410669821.4A patent/CN105678733A/en active Pending
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102663725A (en) * | 2012-03-05 | 2012-09-12 | 西北工业大学 | A visible light and SAR image registration method based on linear features and control points |
Non-Patent Citations (8)
| Title |
|---|
| YAHONG ZHANG 等: "An Automatic Registration Algorithm of Infrared and Visible Images Based on Hybrid Image Features", 《INTERNATIONAL SYMPOSIUM ON OPTOELECTRONIC TECHNOLOGY AND APPLICATION 2014》 * |
| ZHIHENG WANG 等: "HLD: A robust descriptor for line matching", 《2009 11TH IEEE INTERNATIONAL CONFERENCE ON COMPUTER-AIDED DESIGN AND COMPUTER GRAPHICS》 * |
| 刘中杰 等: "基于控制线方法的机载SAR和可见光图像匹配应用研究", 《航空学报》 * |
| 宋莹莹 等: "一种基于形状信息描述的图像特征匹配算法", 《黑龙江科技学院学报》 * |
| 张姣 等: "线特征和SIFT点特征结合的异源图像匹配算法", 《激光与红外》 * |
| 王旭光 等: "Harris相关与特征匹配", 《模式识别与人工智能》 * |
| 艾锐 等: "基于直线映射的红外与可见光图像自动配准算法", 《信息与控制》 * |
| 谢辉 等: "基于局部结构特征的红外与可见光图像匹配", 《计算机工程》 * |
Cited By (22)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108205645A (en) * | 2016-12-20 | 2018-06-26 | 中国科学院沈阳自动化研究所 | A kind of reference map quality evaluating method of heterologous image matching system |
| CN108205645B (en) * | 2016-12-20 | 2020-04-03 | 中国科学院沈阳自动化研究所 | Reference image quality evaluation method of heterogeneous image matching system |
| CN107464257A (en) * | 2017-05-04 | 2017-12-12 | 中国人民解放军军械工程学院 | Wide baseline matching process and device |
| CN107464257B (en) * | 2017-05-04 | 2020-02-18 | 中国人民解放军陆军工程大学 | Wide base line matching method and device |
| CN110189284A (en) * | 2019-05-24 | 2019-08-30 | 南昌航空大学 | A kind of infrared and visible light image fusion method |
| CN110992407A (en) * | 2019-11-07 | 2020-04-10 | 武汉多谱多勒科技有限公司 | Infrared and visible light image matching method |
| CN110992407B (en) * | 2019-11-07 | 2023-10-27 | 武汉多谱多勒科技有限公司 | Infrared and visible light image matching method |
| CN111223134B (en) * | 2020-01-19 | 2023-04-14 | 北京航空航天大学 | Feature vector construction method of linear features, feature matching method and computing equipment |
| CN111223134A (en) * | 2020-01-19 | 2020-06-02 | 北京航空航天大学 | Feature vector construction method of linear features, feature matching method and computing equipment |
| CN111354018B (en) * | 2020-03-06 | 2023-07-21 | 合肥维尔慧渤科技有限公司 | Object identification method, device and system based on image |
| CN111354018A (en) * | 2020-03-06 | 2020-06-30 | 合肥维尔慧渤科技有限公司 | Object identification method, device and system based on image |
| CN112487233A (en) * | 2020-11-27 | 2021-03-12 | 重庆邮电大学 | Infrared and visible light image retrieval method based on feature decoupling |
| CN112487233B (en) * | 2020-11-27 | 2022-07-12 | 重庆邮电大学 | Infrared and visible light image retrieval method based on feature decoupling |
| CN113095385B (en) * | 2021-03-31 | 2023-04-18 | 安徽工业大学 | Multimode image matching method based on global and local feature description |
| CN113095384A (en) * | 2021-03-31 | 2021-07-09 | 安徽工业大学 | Remote sensing image matching method based on context characteristics of straight line segments |
| CN113095385A (en) * | 2021-03-31 | 2021-07-09 | 安徽工业大学 | Multimode image matching method based on global and local feature description |
| CN113221923A (en) * | 2021-05-31 | 2021-08-06 | 西安电子科技大学 | Feature decomposition method and system for multi-mode image block matching |
| CN113221923B (en) * | 2021-05-31 | 2023-02-24 | 西安电子科技大学 | Feature decomposition method and system for multi-mode image block matching |
| CN113792788A (en) * | 2021-09-14 | 2021-12-14 | 安徽工业大学 | Infrared and visible light image matching method based on multi-feature similarity fusion |
| CN113792788B (en) * | 2021-09-14 | 2024-04-16 | 安徽工业大学 | Infrared and visible light image matching method based on multi-feature similarity fusion |
| CN116758067A (en) * | 2023-08-16 | 2023-09-15 | 梁山县成浩型钢有限公司 | Metal structural member detection method based on feature matching |
| CN116758067B (en) * | 2023-08-16 | 2023-12-01 | 梁山县成浩型钢有限公司 | Metal structural member detection method based on feature matching |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105678733A (en) | Infrared and visible-light different-source image matching method based on context of line segments | |
| CN104536009B (en) | Above ground structure identification that a kind of laser infrared is compound and air navigation aid | |
| CN104167003A (en) | Method for fast registering remote-sensing image | |
| CN103593849B (en) | Method for quickly recognizing and tracking image sequence oval artificial target points | |
| CN109543535A (en) | Three-dimensional refers to vena characteristic extracting method and its matching process | |
| CN105046271A (en) | MELF (Metal Electrode Leadless Face) component positioning and detecting method based on match template | |
| US9697438B2 (en) | Image description and image recognizable method | |
| CN105184801A (en) | Optical and SAR image high-precision registration method based on multilevel strategy | |
| CN103886611A (en) | Image matching method suitable for automatically detecting flight quality of aerial photography | |
| CN103714547B (en) | Image registration method combined with edge regions and cross-correlation | |
| CN104268853A (en) | Infrared image and visible image registering method | |
| CN103295232A (en) | SAR (specific absorption rate) image registration method based on straight lines and area | |
| CN107180436A (en) | A kind of improved KAZE image matching algorithms | |
| CN101650784A (en) | Method for matching images by utilizing structural context characteristics | |
| CN106023187A (en) | Image registration method based on SIFT feature and angle relative distance | |
| CN102915441A (en) | Method for detecting unstructured road on basis of Haar texture | |
| CN103839262A (en) | SAR image registration method based on straight lines and FFT | |
| CN103679720A (en) | Fast image registration method based on wavelet decomposition and Harris corner detection | |
| CN105184786A (en) | Floating-point-based triangle characteristic description method | |
| CN105809678B (en) | A kind of line segment feature global registration method between two views under short base line condition | |
| Xu et al. | Ring++: Roto-translation-invariant gram for global localization on a sparse scan map | |
| CN104282001A (en) | Method for enhancing image feature two-value descriptor performance | |
| CN103336964A (en) | SIFT image matching method based on module value difference mirror image invariant property | |
| CN102902973A (en) | Extraction method of image characteristic with rotation invariance | |
| CN101819636A (en) | Irregular area automatic matching method in the digital picture |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20160615 |
|
| WD01 | Invention patent application deemed withdrawn after publication |