CN105469417A - MSER (Maximally Stable Extremal Regions) + PSO (Particle Swarms Optimization) step-by-step multi-sensor image registration algorithm - Google Patents
MSER (Maximally Stable Extremal Regions) + PSO (Particle Swarms Optimization) step-by-step multi-sensor image registration algorithm Download PDFInfo
- Publication number
- CN105469417A CN105469417A CN201511026534.2A CN201511026534A CN105469417A CN 105469417 A CN105469417 A CN 105469417A CN 201511026534 A CN201511026534 A CN 201511026534A CN 105469417 A CN105469417 A CN 105469417A
- Authority
- CN
- China
- Prior art keywords
- image
- registration
- pso
- mser
- estimated value
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20016—Hierarchical, coarse-to-fine, multiscale or multiresolution image processing; Pyramid transform
Abstract
The invention belongs to the technical field of image registration and relates to a MSER (Maximally Stable Extremal Regions) + PSO (Particle Swarms Optimization) multi-sensor image registration algorithm. In comparison with the traditional grayscale-based registration algorithm, the method participating in searching for the best matching point is carried out step by step rather than traversing all pixels. Firstly, a feature point-based method is used for initial registration, and the gap between a to-be-registered image and a reference image is reduced; then a method of two times of PSO is used for fine registration, the two times of searching have different starting points, and the matched best point can be obtained after two times of PSO. As an edge area image participates in operation, image data participating in operation actually is reduced to 8% to 25% of the original data, the registration time is shorted to be about 15% to 40% of the original time, and as two times of PSO are adopted, the registration precision can achieve a sub pixel level.
Description
Technical field
The invention belongs to image registration techniques field, relate to the allos image registration algorithm of a kind of MSER+PSO.
Background technology
Image registration is widely used in the practical problemss such as multisource data fusion, image mosaic, object variations detection, target identification, but the precision of image registration and real-time cannot be solved all the time effectively, especially the registration of allos image is the difficult point in registration field always.
The method of image registration is generally divided into two classes, the method for feature based and the method based on region.The method speed of feature based is fast, but registration accuracy is low, only has unique point pixel to participate in calculating, is generally used for the registration of same source images; And high based on the method precision in region, but registration speed is slow, is used for the registration of allos image.And based on the method in region, as the method for registering based on gray scale does not need to do feature extraction to image usually, but directly utilize the half-tone information of entire image, set up the similarity measurement (as mutual information, cross-correlation etc.) between two width images, then adopt certain chess game optimization method, find the parameter value of the transformation model making similarity measure values maximum or minimum.Because make use of whole gradation of image information, thus estimate precision and robustness higher.But in this approach, the data of entire image all will participate in computing, therefore its calculated amount is very large, and the registration time is long.The method and access that above problem makes the registration problems of allos image still need.The method of current head it off has: 1. when adopting the method for registering based on gray scale, improve certain chess game optimization method, as improved ant group algorithm, genetic algorithm or particle cluster algorithm, find the parameter value of the transformation model making similarity measure values maximum or minimum, this method can improve the real-time of registration to a certain extent; 2. for breaking through the limitation of a certain or a certain class methods, two kinds of features or two kinds of methods are combined.As SIFT descriptor is carried out major component registration by PCA, 1. ratio method has more real-time; Or successively adopt the method for unique point and the method registration of ant group optimization, also obtain ratio method 1. better real-time.Two class methods improve the real-time of registration all to a certain extent, but the precision of registration need to be improved further.
Summary of the invention
Given this, the present invention proposes a kind of MSER (MaximallyStableExtremalRegions, MSER)+PSO (ParticleSwarmsOptimization, PSO) multiple step format allos image registration algorithm, mainly for the situation between reference picture and image subject to registration being affined transformation (namely there is translation, rotation).
In order to solve the problems of the technologies described above, the present invention is achieved in that
(1) maximum stable extremal region is obtained with reference to image R and image S subject to registration by MSER method;
(2) only extract minutiae in the maximum stable extremal region that step (1) obtains, and selected distance center of gravity n farthest point is as unique point, n>3; Then adopt the method for registering of feature based to carry out registration, obtain first registering images S1;
(3) the first registering images S1 obtained with reference to image R and step (2) carries out 1/2 down-sampling;
(4) image after step (3) down-sampling is passed through primary particle group hunting PSO, the initial value diameter of Spherical Volume A of population search
0=[0,0,0], search radius is r=5, obtains time figure of merit
wherein
representing the estimated value of first horizontal shift, is just to the right;
representing the estimated value of first perpendicular displacement, is just downwards;
represent first anglec of rotation estimated value, and with image geometry center for initial point, just clockwise turn to.(5) binary edge map is extracted to the first registering images S1 that reference image R and step (2) obtain; Then find out the pixel of isolated edge d<D, D is 5, obtains fringe region mask, and extract the fringe region of original image with the fringe region mask obtained, registration image used is fringe region image;
(6) offspring group hunting is passed through by two breadths edge area images in (5), the initial value diameter of Spherical Volume of current population search
r=5, obtains optimal value
wherein
represent the estimated value of terminal level displacement,
representing the estimated value of final perpendicular displacement, is just upwards;
represent final anglec of rotation estimated value.Direction definition is with (4).
Beneficial effect:
The starting point of multiple step format Optimizing Search algorithm of the present invention is in conjunction with feature based and the advantage based on gray scale two kinds of method for registering.First adopt at the beginning of the method for registering of distinguished point based and join, reduce the change of scale between image subject to registration and reference picture as far as possible, and obtain the image S1 after first registration; Again using S1 as image subject to registration, and reference image R adopts the method for particle group optimizing to carry out carefully joining for twice.It is fast and based on the high feature of grey scale accuracy that this method takes full advantage of feature based method speed, improves the precision of registration while improving registration speed.
Visible, the present invention compares with traditional registration Algorithm based on gray scale, and its method participating in finding optimal match point is not by traveling through all pixels, but carry out step by step.First utilize at the beginning of the method for distinguished point based and join, reduce the gap of image subject to registration and reference picture, and then the mode of PSO (particle group optimizing) at twice is carefully joined, but the starting point of twice search is different, can obtain the optimum point of mating after offspring group hunting.Is fringe region image due to what participate in computing, the view data of actual participation computing is reduced to original about 8% ~ 25%, and time of registration also shortens to original about 15% ~ 40%.Owing to have employed twice population search (PSO), the precision of registration can reach sub-pixel.
Accompanying drawing explanation
Fig. 1 tradition is based on gray scale registration Algorithm flow process
Fig. 2 is the schematic flow sheet of MSER+PSO allos image registration algorithm.
Fig. 3 is the spherical distribution space schematic diagram of PSO initial value.
Fig. 4 is edge mask leaching process
Embodiment
According to the feature of image registration algorithm, algorithm is decomposed into 6 steps, specifically:
(1) maximum stable extremal region is obtained with reference to image R and image S subject to registration by MSER method; MSER and maximum stable extremal region are a kind of methods of maturation.
(2) only extract minutiae in the maximum stable extremal region that step (1) obtains, and selected distance center of gravity n (n>3 farthest, specifically self-defined) individual point is as unique point, then adopt the method for registering of feature based to carry out registration, obtain first registering images S1.
(3) the first registering images S1 obtained with reference to image R and step (2) carries out 1/2 down-sampling.Because image becomes original 1/4 after down-sampling, the time of a PSO (particle group optimizing) can be reduced.
(4) image after step (3) down-sampling is passed through a PSO (initial value diameter of Spherical Volume A
0=[0,0,0], r=5) after, obtain time figure of merit
(5) binary edge map is extracted to the first registering images S1 that reference image R and step (2) obtain; Then find out the pixel of isolated edge d<D, obtain fringe region mask, extract the fringe region of original image with the fringe region mask obtained, registration image used is fringe region image.Edge image is the part that entire image frequency is the highest, contains detailed image structure information, contributes to the precision improving image registration
(6) secondary PSO (initial value diameter of Spherical Volume is passed through by two breadths edge area images in (5)
R=5) optimal value is obtained
Claims (1)
1. an allos image registration algorithm of MSER+PSO, is characterized in that step is as follows:
(1) maximum stable extremal region is obtained with reference to image R and image S subject to registration by MSER method;
(2) only extract minutiae in the maximum stable extremal region that step (1) obtains, and selected distance center of gravity n farthest point is as unique point, n>3; Then adopt the method for registering of feature based to carry out registration, obtain first registering images S1;
(3) the first registering images S1 obtained with reference to image R and step (2) carries out 1/2 down-sampling;
(4) image after step (3) down-sampling is passed through primary particle group hunting PSO, the initial value diameter of Spherical Volume A of population search
0=[0,0,0], search radius is r=5, obtains time figure of merit
wherein
representing the estimated value of first horizontal shift, is just to the right;
representing the estimated value of first perpendicular displacement, is just downwards;
represent first anglec of rotation estimated value, and with image geometry center for initial point, just clockwise turn to;
(5) binary edge map is extracted to the first registering images S1 that reference image R and step (2) obtain; Then find out the pixel of isolated edge d<D, D is 5, obtains fringe region mask, and extract the fringe region of original image with the fringe region mask obtained, registration image used is fringe region image;
(6) offspring group hunting is passed through by two breadths edge area images in step (5), the initial value diameter of Spherical Volume of current population search
r=5, obtains optimal value
wherein
represent the estimated value of terminal level displacement,
representing the estimated value of final perpendicular displacement, is just upwards;
represent final anglec of rotation estimated value; Direction definition is with (4).
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201511026534.2A CN105469417A (en) | 2015-12-30 | 2015-12-30 | MSER (Maximally Stable Extremal Regions) + PSO (Particle Swarms Optimization) step-by-step multi-sensor image registration algorithm |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201511026534.2A CN105469417A (en) | 2015-12-30 | 2015-12-30 | MSER (Maximally Stable Extremal Regions) + PSO (Particle Swarms Optimization) step-by-step multi-sensor image registration algorithm |
Publications (1)
Publication Number | Publication Date |
---|---|
CN105469417A true CN105469417A (en) | 2016-04-06 |
Family
ID=55607076
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201511026534.2A Pending CN105469417A (en) | 2015-12-30 | 2015-12-30 | MSER (Maximally Stable Extremal Regions) + PSO (Particle Swarms Optimization) step-by-step multi-sensor image registration algorithm |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN105469417A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111738993A (en) * | 2020-06-05 | 2020-10-02 | 吉林大学 | G-W distance-based ant colony graph matching method |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101216939A (en) * | 2008-01-04 | 2008-07-09 | 江南大学 | A multi-resolution medical image registration method based on quantum behaviors particle swarm algorithm |
CN102509303A (en) * | 2011-11-22 | 2012-06-20 | 鲁东大学 | Binarization image registration method based on improved structural similarity |
US8645294B1 (en) * | 2004-02-03 | 2014-02-04 | Hrl Laboratories, Llc | Method for image registration utilizing particle swarm optimization |
-
2015
- 2015-12-30 CN CN201511026534.2A patent/CN105469417A/en active Pending
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US8645294B1 (en) * | 2004-02-03 | 2014-02-04 | Hrl Laboratories, Llc | Method for image registration utilizing particle swarm optimization |
CN101216939A (en) * | 2008-01-04 | 2008-07-09 | 江南大学 | A multi-resolution medical image registration method based on quantum behaviors particle swarm algorithm |
CN102509303A (en) * | 2011-11-22 | 2012-06-20 | 鲁东大学 | Binarization image registration method based on improved structural similarity |
Non-Patent Citations (4)
Title |
---|
YE ZHANG: "Particle Swarm Optimization with Powell"s Direction Set Method for Remote Sensing Image Registration", 《2009 FIFTH INTERNATIONAL CONFERENCE ON NATURAL COMPUTATION》 * |
杜玉平: "关于粒子群算法改进的研究", 《中国优秀硕士学位论文全文数据库基础科学辑》 * |
程丹 等: "基于MSER的SAR影像配准算法", 《测绘通报》 * |
章学静: "像素级图像增强及配准算法研究", 《中国博士学位论文全文数据库信息科技辑》 * |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111738993A (en) * | 2020-06-05 | 2020-10-02 | 吉林大学 | G-W distance-based ant colony graph matching method |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN103400384B (en) | The wide-angle image matching process of calmodulin binding domain CaM coupling and some coupling | |
KR101856401B1 (en) | Method, apparatus, storage medium, and device for processing lane line data | |
CN105631880B (en) | Lane line dividing method and device | |
Zhang et al. | Ripple-GAN: Lane line detection with ripple lane line detection network and Wasserstein GAN | |
CN107767382A (en) | The extraction method and system of static three-dimensional map contour of building line | |
CN110136159B (en) | Line segment extraction method for high-resolution remote sensing image | |
CN106067003A (en) | Road vectors tag line extraction method in a kind of Vehicle-borne Laser Scanning point cloud | |
CN110379004B (en) | Method for classifying ground features and extracting single objects of oblique photography results | |
CN104200461A (en) | Mutual information image selected block and sift (scale-invariant feature transform) characteristic based remote sensing image registration method | |
CN108256467B (en) | Traffic sign detection method based on visual attention mechanism and geometric features | |
CN110136174B (en) | Target object tracking method and device | |
CN113628291B (en) | Multi-shape target grid data vectorization method based on boundary extraction and combination | |
CN113673541B (en) | Image sample generation method for target detection and application | |
CN111047603B (en) | Aerial image hybrid segmentation algorithm based on novel Markov random field and region combination | |
CN113449594A (en) | Multilayer network combined remote sensing image ground semantic segmentation and area calculation method | |
CN104063711A (en) | Corridor vanishing point rapid detection algorithm based on K-means method | |
CN107133623A (en) | A kind of pointer position accurate detecting method positioned based on background subtraction and the center of circle | |
CN104851095A (en) | Workpiece image sparse stereo matching method based on improved-type shape context | |
CN103955950A (en) | Image tracking method utilizing key point feature matching | |
CN109389165A (en) | Oil level gauge for transformer recognition methods based on crusing robot | |
CN115131363A (en) | Positioning method and device based on semantic information and terminal equipment | |
CN105469417A (en) | MSER (Maximally Stable Extremal Regions) + PSO (Particle Swarms Optimization) step-by-step multi-sensor image registration algorithm | |
CN109816710B (en) | Parallax calculation method for binocular vision system with high precision and no smear | |
Li et al. | Automatic Road Extraction from High-Resolution Remote Sensing Image Based on Bat Model and Mutual Information Matching. | |
CN106056575A (en) | Image matching method based on object similarity recommended algorithm |
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 | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20160406 |