CN109785370A - A kind of weak texture image method for registering based on space time series model - Google Patents
A kind of weak texture image method for registering based on space time series model Download PDFInfo
- Publication number
- CN109785370A CN109785370A CN201811517008.XA CN201811517008A CN109785370A CN 109785370 A CN109785370 A CN 109785370A CN 201811517008 A CN201811517008 A CN 201811517008A CN 109785370 A CN109785370 A CN 109785370A
- Authority
- CN
- China
- Prior art keywords
- image
- registration
- feature
- reference picture
- model
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Abstract
A kind of weak texture image method for registering based on space time series model proposed by the present invention, comprising the following steps: obtain reference picture and image subject to registration；The characteristic point of use space temporal model method detection reference picture and image subject to registration；Determine the direction of characteristic point, construction describes son comprising the compound characteristics point of Gradient direction information；Feature Points Matching is carried out to reference picture and image subject to registration by signature search；Erroneous matching pair is rejected, optimal matching points and transformation model are obtained；Image subject to registration is converted by transformation model, obtains registration result.For the present invention compared with region, the present invention replaces direct gray scale operation with characteristics of image, has many advantages, such as that characteristic method is high-efficient, accurate；Compared with characteristic method, present invention reduces algorithms to the degree of dependence of notable feature, does not need manually to add supplemental characteristic, avoids part from scratching, improve the robustness of algorithm, autoregistration may be implemented.
Description
Technical field
The present invention relates to image registration fields more particularly to a kind of weak texture image based on space time series model to match
Quasi- method.
Background technique
Image registration is two width or several figures for acquiring same scene in different time, different perspectives, different sensors
As the process being overlapped mutually, it realizes the alignment on reference picture and image geometry meaning subject to registration.The application of image registration
Range is very extensive, including machine vision, three-dimensional reconstruction, remote sensing image processing, target classification and retrieval, image understanding and fusion
Deng.Method for registering images includes being registrated method based on region, based on line feature or region feature and based on point feature.
Region registration method directly utilizes the similarity degree of the grayscale information calculating image of image, treatment process are as follows: first
The similarity metric function for being suitble to feature of image is selected, then in selected geometric transformation model parameter spatial class, according to certain
Search strategy scan for, finding can enable similarity reach maximum geometric transformation parameter.Such methods be by
It searches in registration image and is registrated with the maximum region of reference picture similarity, be not related to the extraction of feature, adopted in matching
Block is used to carry out matching estimation as registration unit, and the selection of similarity metric function is to determine the accuracy and effect of template registration
The key factor of rate height.
The Yin Sumin of Jiangsu University, which is proposed, a kind of realizes the on-line automatic of cylinder body based on the detection algorithm of template matching
Detection.Experiment is realized in the specific zonule of engine cylinder body using the improved sequential similarity algorithm of coarse-fine combination search strategy
(Yin Sumin, Bao Hongli, Ji Binbin wait to pass based on engine cylinder body defects detection [J] of zonule template matching to defects detection
Sensor and micro-system, 2012 (06): 143-145.).The Wu Xiaojun etc. of Shenzhen Space Technology Innovation Academe proposes a kind of fast
There is situations such as displacement, rotation, scaling, partial occlusion, Ke Yiying for target image in fast high-precision geometric templates method for registering
Occasion (the quick high accuracy geometric templates match party that a kind of band of Wu Xiaojun is rotated, scaled demarcated and identified for machine vision
Method: China, CN2016102093086 [P] .2016-09-07.).Wang Monan of Harbin University of Science and Technology et al. establishes a set of
Medical image atlas image library, since there is clearly resolution using the gray level image of CT shooting, according to target image label
With the map label in image library, in conjunction with the grayscale information and gradient information of image tag, by best atlas image tag fusion
At piece image as registration template, effectively raising the accuracy of registration, (Wang Monan, Li Pengcheng, Jing Juntong one kind are based on
The Medical Image Registration Algorithm of multichannel chromatogram tag fusion: China, CN107093190 [P] .2017.08.25.).
Region is registrated extraction of the method independent of feature, but carries out global operations to whole sub-picture, and principle is simple, is easy to
It realizes.In recent years, for the improvement direction of this kind of algorithm, the building of similarity measure and the optimization of search strategy are concentrated mainly on
On.But the building by improving similarity measure, all kinds of algorithms can not still solve the information content excessive caused algorithm time
The high problem of complexity.By improved search strategy, search range is successively reduced, is improved the work of efficiency to a certain extent
With.But it is different from clearly demarcated gray level image clear in medical image, the applications such as large-scale part measurement often occur big
The weak texture region of area.Gray value is close everywhere for weak texture region, there is during rough registration and largely mismatches standard, leads
Cause algorithm failure.In addition, picture noise is inevitably present, so that image grayscale changes, to influence the essence of registration
Degree.
Point feature is registrated method, generally using angle point or characteristic point as the basic unit of images match, by from benchmark image
Corresponding relationship between image subject to registration estimates the value of geometric transformation model and parameter between the two.Selection is closed first
Suitable feature vertex type, carries out the detection of characteristic point, then matches the characteristic point in two images, chooses corresponding geometric transformation
Model simultaneously carries out parameter estimation, converts finally by image and completes registration.The description method of local image characteristics is to determine point
The key of feature registration method accuracy and efficiency.
Liu Bo of Harbin University of Science and Technology et al. is proposed based on pyramid structure and the improved SIFT of feature vertex neighborhood
Algorithm is carried out denoising, the reduction grayscale information of image by the original image obtained to imaging system, it is extensive to remove unrelated information
Multiple useful information, the characteristic point quantity obtained after feature extraction can greatly reduce, thus improve registration speed (Liu Bo,
A kind of improved image registration algorithm based on SIFT feature of Pang Ying: China, CN107886530A [P] .2018.04.06.).
Sang Hongshi of the Central China University of Science and Technology etc. extracts a certain number of characteristic points, first when carrying out feature detection to image with characteristic point
Coordinate and its neighborhood generate mask images, carry out second of feature detection, the feature that will be detected twice using mask images
Point merges, and exports as final feature testing result.The positional relationship between characteristic point is utilized, it is weaker in characteristic strength
Region still retains a kind of a certain number of characteristic points (image registration side based on the detection of feature twice Sang Hongshi, Dong Tong, Hu Peng
Method: China, CN108182700A [P] .2018.06.19.).Zheng Hong of Wuhan University et al. proposes a kind of towards the more of workpiece
Dimensional characteristics method for registering, by obtaining the ORB characteristic point of template workpiece picture and workpiece picture to be positioned, in conjunction with based on k-
The various dimensions registration Algorithm of means cluster constraint, improves the robustness and positioning accuracy (Zheng Hong, Zheng Chaohui mono- of matching algorithm
The various dimensions feature registration method that kind is positioned towards workpiece: China, CN108502533A.2018.9.11.).
It is very widely used that point feature is registrated method.Point feature is richly stored with image structure information, and registration has aobvious
The image of point feature is write, aforementioned several method has higher robustness and registration accuracy.However, being surveyed in large-scale part vision
In amount, usually only machining is formed by more microcosmic " trace " in topography, and there is no significant point features to exist,
It manually the supplementary means such as portrays if taken, workpiece surface may be scratched, be also unfavorable for realizing on-line checking.
Line feature or region feature are registrated method, and the general discontinuity using image local gray scale determines that the edge line of image is special
Sign, is related to asking the first derivative or second dervative of image grayscale function.Firstly, image preprocessing is carried out as needed, by figure
Picture filtering and image enhancement make the selection of compromise between enhancing edge and abating noises, then anxious using image border gray scale
The characteristics of drastic change, detects edge feature.It is completed finally, the corresponding relationship according to picture edge characteristic restores geometric transformation model
Registration based on line feature.In addition, region feature registration method is dependent on edge detection as a result, being enclosed on its basis with edge contour
At the features such as centroid, the center of gravity of enclosed region as registration foundation.
Xian Electronics Science and Technology University Wang Shuan et al. is by SAR image Canny operator extraction edge, according to edge fitting
Detection straight line obtains the rotation parameter between image, then carries out Fourier transformation to postrotational image subject to registration, calculates image
Between horizontal translation amount and vertical translation amount complete SAR image registration (Wang Shuan, Jiao Licheng, Zhang Nan, Liu Kun, Marvin's duckweed, horse
Jingjing, a kind of SAR image registration method based on straight line and FFT of Zhang Tao, Liu Chuan: China,
CN103839262A.2014.06.04.).The Zhang Yanning of Northwestern Polytechnical University is by extracting marking area in non-homogeneous image, benefit
The feature that marking area is described with Zernike rotation invariant moment improves precision (Zhang Yanning, the Chen Yan of non-homogeneous image registration
Good, Zhang Xiuwei, yellow plum is tender, the visible light based on the marking area feature and edge degree-infrared image registration side Li Fei, Fu Zhipeng
Method: China, CN106447704A.2017.02.22.).Li Huiqi of Beijing Institute of Technology etc. extracts X by Canny algorithm and penetrates
The profile information on the head of line chart piece and photochrome selects fixed reference feature point auxiliary to carry out rough registration, recycles consistent point drift
It moves algorithm progress Curve Matching and has obtained blending image (Li Huiqi, Wang Shumeng one kind of X-ray picture and photochrome superposition
X-ray film and photochrome method for registering based on contour feature: China, CN107240128A.2017.10.10.).
Line/region feature registration method first has to carry out edge detection, then is filtered out and met the requirements by some restrictive conditions
Line segment as local feature.Natural image usually has more significant line/region feature, so line/region feature method for registering tool
There are higher registration accuracy and robustness.However, the processing " trace " contained in part image is usually fainter, and image
Gray scale difference is also little, and even with preprocessing means such as image enhancements, edge is still not easy to detect.In addition, processing " trace " tool
There is apparent periodicity, then only describing son using edge self information construction feature point, may cause registration be cannot achieve.
Summary of the invention
The present invention is exactly in view of the deficienciess of the prior art, provide a kind of using space time series model detection figure
As characteristic point, the weak texture image registration based on space time series model without notable feature image feature extraction problem is solved
Method.
To solve the above problems, the technical solution used in the present invention is as follows:
A kind of weak texture image method for registering based on space time series model, method includes the following steps:
S1: the weak texture image without significant point or line feature is obtained as reference picture, obtains another secondary and reference picture
There is the weak texture image of certain degree of overlapping as image subject to registration；
S2: the characteristic point of use space temporal model method detection reference picture and image subject to registration；
S3: determining the direction of characteristic point, and construction describes son comprising the compound characteristics point of Gradient direction information；
S4: Feature Points Matching is carried out to reference picture and image subject to registration by signature search；
S5: rejecting erroneous matching pair, obtains optimal matching points and transformation model；
S6: image subject to registration is converted by transformation model, obtains registration result.
Preferably, specific step is as follows by step S2:
S21: each pixel on reference picture is selected one by one；
S22: taking image block from reference picture around selected pixel, establishes space temporal model, solving model parameter；
S23: model built applicability is examined using AIC criterion；
S24: based on types of models parameter and detection applicability as a result, describing son to selected pixel feature points.
Preferably, step S include thes steps that carrying out conspicuousness detection S25 to characteristic point, and steps are as follows by step S25:
Reference picture or image subject to registration are divided into several image blocks；
Successively calculate other each pixels in feature point description and current block of selecting pixel in each image block
Feature point description Euclidean distance, sum after Euclidean distance value is weighted, obtain weighted sum；
When weighted sum is greater than threshold value, determines that selected pixel is the characteristic point of conspicuousness, remain and match for image
Standard abandons it, is not used in image registration conversely, selected pixel is not the characteristic point of conspicuousness.
Preferably, specific step is as follows by step S3:
Gradient-norm and the direction for calculating selected characteristic point, with each pixel in the selected feature vertex neighborhood of statistics with histogram
The gradient direction and amplitude of point, histogram peak direction indicates the principal direction of characteristic point, based on constructing in principal direction and step S2
Feature point description son, construction compound characteristics point description son.
Preferably, in step s 4, using KD tree or quickly, nearest neighbor algorithm or K- nearest neighbor algorithm carry out signature search,
Find the corresponding relationship between reference picture and image characteristic point subject to registration.
Preferably, in step s 5, purification processes are carried out to the characteristic point obtained in S4 using random sampling unification algorism,
Reject erroneous matching pair.
Compared with prior art, implementation result of the invention is as follows by the present invention:
Compared with field method, the present invention replaces direct gray scale operation with characteristics of image, high-efficient, accurate etc. with characteristic method
Advantage；Compared with characteristic method, present invention reduces algorithms to the degree of dependence of notable feature, does not need manually to add auxiliary spy
Sign, avoids part from scratching, improves the robustness of algorithm, autoregistration may be implemented.
Detailed description of the invention
Fig. 1 is the forecast window of space temporal model of the invention.
Fig. 2 is the flow chart of temporal model registration image in space of the invention.
Fig. 3 is the flow chart of temporal model detection characteristic point in space of the invention.
Specific embodiment
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Describe to whole property.Certainly, described embodiment is a part of the embodiment of the present invention, instead of all the embodiments.
Referring to Fig.1, Fig. 1 is the forecast window of space temporal model of the invention, in figure: 1, with reference to the ash of (or wait forecast)
Image is spent, size is m × n-pixel；2, pixel to be forecast；3, known (or being history) pixel.
In conjunction with Fig. 2 and Fig. 3, a kind of weak texture image registration side based on space time series model proposed by the present invention
Method, following steps:
(1) reference picture and image subject to registration are obtained
The weak texture image without significant point or line feature is obtained as reference picture, obtaining another secondary and reference picture has one
The weak texture image of degree of overlapping is determined as image subject to registration, and reference picture and image subject to registration are gray level images.
(2) characteristic point of the detection of use space temporal model method reference picture and image subject to registration
(1) construction feature point description
Space temporal model G (r；w；w1,w2,...,wr) there are various types, it depends on the type parameter of model, including
Model order r, it returns window type w and returns window size wi(i=1,2 ..., r).Order r takes positive integer, r=1 table
Show space temporal model be it is linear, 1 representation space temporal model of r > is nonlinear.W takes 1,2,3,4 different values, indicates to adopt
Recurrence window type is non-causal type respectively, partly because of fruit type, because of fruit type, by force because fruit type returns window.Black dot indicates
To prediction pixel point, (or being history) pixel known to the expression of grey dot, grey and black pixel point are encircled a city from left to right
Region be respectively non-causal type, half that window size is 2 because of fruit type, because of fruit type, by force because fruit type returns window.It is various types of
The number of space temporal model, modeling parameters is different, and indicates G (r with a column vector α；w；w1,w2,...,wr) model
Modeling parameters.For the applicability for examining model built, model testing is carried out with AIC criterion, inspection result is expressed as AIC (r).
In the optional pixel (u, v) of reference picture (or image subject to registration), its gray value is f (u, v).
Select model G (1；1；1), i.e., the order of model is 1, and returning window type is that non-causal type returns window, is returned
The size of window is 1.
From original image clipping image block B1(u, v)={ (s, t) ∈ z2:u-1≤s≤u+4,v-1≤t≤v+2}；
Define plane point set B2(p, q)={ (p, q) ∈ Z2: 1≤p≤6,1≤q≤4 }, that is, B can be used2(p, q) is
Indicate image block B1The pixel coordinate of (u, v) indicates the gray value of pixel with f (p, q), establishes space temporal model:
F (h, l)=a (h, l) α+ε (h, l)
In formula, h ∈ Z and 2≤h≤5, l ∈ Z and 2≤l≤3；α is G (1；1；1) model modeling parameter is one 8 dimension column
Vector；A (h, l) is one 8 dimension row vector, and element is f (h ', l '), (h ', l ') ∈ N8(h,l)。
H and l take different value, total to list 8 formulas, are write this 8 formulas as matrix A α=b form, A=in formula a (h,
L) }, b=(f (h, l))T。
Formula has 8 unknown parameters, therefore more than simultaneous 8 formulas, can be solved to obtain modeling parameters with least square method
α.Using α as iterative initial value, estimated to update model parameter with GM (GeneralizedM-estimator)According to changing for GM estimation
For known to update mechanismThere is better noiseproof feature than α.
The variance of computation modeling residual errorWhereinUtilize being applicable in for AIC criterion testing model
Property,r1It is model order.
Using modeling parameters and model applicability examine as a result, one 9 dimensional feature vector of construction
Select model G (1；1；2), i.e., the order of model is 1, and returning window type is that non-causal type returns window, is returned
The size of window is 2.
From original image clipping image block B1(u, v)={ (s, t) ∈ z2:u-2≤s≤u+7,v-2≤t≤v+5}；
Define plane point set B2(p, q)={ (p, q) ∈ Z2: 1≤p≤10,1≤q≤8 }, that is, B can be used2(p,q)
It is to indicate image block B1The pixel coordinate of (u, v) indicates the gray value of pixel with f (p, q), establishes space temporal model:
A α+ε=b
In formula, A={ a (h, l) }, α are modeling parameters (24 dimensional vector), b=(f (h, l))T, h, l ∈ Z and 2≤h≤
7, and 2≤l≤5, a (h, l) they are one 24 dimension row vectors, element is f (h ', l '), (h ', l ') ∈ N24(h,l)。
Least square method can solve to obtain formula parameter alpha, and GM estimation updates model parameter
Calculate the variance of residual errorWhereinAIC criterion testing model applicability, result arer2It is model order.
Using modeling parameters and model applicability examine as a result, 25 dimensional feature vectors of construction
Select model G (2；1；1,1), i.e., the order of model is 2, and returning window type is that non-causal type returns window, is returned
The size for returning window is 1.
From original image clipping image block B1(u, v)={ (s, t) ∈ z2:u-1≤s≤u+7,v-1≤t≤v+7}；
Define plane point set B2(p, q)={ (p, q) ∈ Z2: 1≤p≤9,1≤q≤9 }, that is, B can be used2(p, q) is
Indicate image block B1The pixel coordinate of (u, v) indicates the gray value of pixel with f (p, q), establishes space temporal model:
A α+ε=b
In formula, A={ a (h, l), c (h, l) }, α are modeling parameters (44 dimensional vector), b=(f (h, l))T, h, l ∈ Z and
2≤h≤8, and 2≤l≤8.Wherein, a (h, l) is one 8 dimension row vector, and element is f (h ', l '), (h ', l ') ∈ N8(h,l)。
C (h, l) is one 36 dimension row vector, and the rejecting duplicate keys that are multiplied two-by-two by 8 elements in a (h, l) obtain.
Least square method can solve to obtain formula parameter alpha, and GM estimation updates model parameter
Calculate the variance of residual errorWhereinAIC criterion testing model applicability, result arer3It is model order.
Using modeling parameters and model applicability examine as a result, 45 dimensional feature vectors of construction
Select model G (2；1；2,1), i.e., the order of model is 2, and returning window type is that non-causal type returns window, line
Property return window size be 2, nonlinear regression window size be 1.
From original image clipping image block B1(u, v)={ (s, t) ∈ z2:u-2≤s≤u+9,v-2≤t≤v+9}。
Define plane point set B2(p, q)={ (p, q) ∈ Z2: 1≤p≤12,1≤q≤12 }, that is, B can be used2(p,q)
It is to indicate image block B1The pixel coordinate of (u, v) indicates the gray value of pixel with f (p, q), establishes space temporal model:
A α+ε=b
In formula, A={ a (h, l), c (h, l) }, α are modeling parameters (60 dimensional vector), b=(f (h, l))T, h, l ∈ Z and
3≤h≤10, and the same G (1 of 3≤l≤10, a (h, l)；1；2) model, the same to G (2 of c (h, l)；1；1,1) model.
Least square method can solve to obtain formula parameter alpha, and GM estimation updates model parameter
Calculate the variance of residual errorWhereinAIC criterion testing model applicability, result arer4It is model order.
Using modeling parameters and model applicability examine as a result, 61 dimensional feature vectors of construction
Construction feature point describes sub- F (u, v)=(f1,f2,f3,f4), F (u, v) is the row vector of one 140 dimension.
Change u or v value, reselect pixel, repeats above-mentioned modeling process, construct the feature point description of new pixel
Son.So until traversal entire image, defines feature point description for each pixel.
(2) conspicuousness of characteristic point is detected
Reference picture (or image subject to registration) is divided into several image blocks, k-th of tile size is mk×nk, examine
The conspicuousness of characteristic point in image block, detailed process is as follows:
Optional two characteristic points, calculate Euclidean distance between the two on image block, constitute one with whole calculated results
A column vector dk, then dkDimension is C (mk×nk,2)。
Delete dkIn repeat element (only retain one), then by dkAccording to element descending order arrange, update to
Amount is usedIt indicates, dimension is denoted as ε.
Using elementary function y=f (x) as weight function, meet f (x) monotone decreasing in domain, and f (x) >=1.Meet
The elementary function of above-mentioned condition can be y=aγ(x-ε), 0 < a <, 1,0 < γ < 1 in formula
Some optional pixel (p, q) on image block, feature point description vector are F (p, q).On image block optionally
One pixelIts feature point description vector isIt calculatesD is searched to exist
Middle present position x substitutes into x in function y=f (x), the distance using functional value y as weight, after calculating weighting
ChangeOrValue recalculatesD is searched again to existMiddle present position calculates
One new weight calculates distance after weightingSuch Repeated mk×nk- 1 time, until traversing whole image block, in total
Obtain mk×nk- 1 Weighted distance sums these Weighted distances, and summation then determines that pixel (p, q) is if it is greater than threshold value
Significant characteristic point will use it in image registration, and otherwise just abandon it.
Change p or q value, repeats the above process mk×nkIt is secondary, complete the significance test of whole characteristic points on image block.
Change k value, repeat the above process, completes the significance test of whole characteristic points in entire image.
(3) construction meets feature point description
It is characterized a distribution directional information, to solve rotation image registration problem.Firstly, calculating separately picture according to formula and formula
Plain coordinate is the gradient-norm and direction of (u, v) characteristic point:
Using centered on (u, v), using W/2 as radius (W be space temporal model modeling when window size), use statistics with histogram
The gradient direction and amplitude of each pixel, histogram peak direction represent the principal direction of key point in selected feature vertex neighborhood
β.Feature point description constructed using principal direction information and front constructs compound feature point description
(4) signature search with match；
By signature search and matching, the corresponding relationship between reference picture and image characteristic point subject to registration is found.Feature
It is very high that point describes sub (vector) dimension, to improve search efficiency, reduces memory space complexity, can be used and some calculates quickly fastly
Method, such as KD tree, quick nearest neighbor algorithm, K- nearest neighbor algorithm etc..
(5) erroneous matching pair is rejected, optimal matching points and transformation model are obtained；
The characteristic point obtained using previous step is solved the parameter of geometric transformation model, realizes reference picture to (at least 8 pairs)
With the registration of image subject to registration.Consider that characteristic point to there are error hidings, can carry out at purification thick matching using respective algorithms
Reason, such as random sampling unification algorism (Random Sample Consensus, RANSAC).
(6) image subject to registration is converted by transformation model, obtains registration result.
The foregoing is merely illustrative of the preferred embodiments of the present invention, is not intended to limit the invention, all in essence of the invention
Made any modifications, equivalent replacements, and improvements etc., should all be included in the protection scope of the present invention within mind and principle.
Claims (6)
1. a kind of weak texture image method for registering based on space time series model, which is characterized in that this method includes following
Step:
S1: the weak texture image without significant point or line feature is obtained as reference picture, obtaining another secondary and reference picture has one
The weak texture image of degree of overlapping is determined as image subject to registration；
S2: the characteristic point of use space temporal model method detection reference picture and image subject to registration；
S3: determining the direction of characteristic point, and construction describes son comprising the compound characteristics point of Gradient direction information；
S4: Feature Points Matching is carried out to reference picture and image subject to registration by signature search；
S5: rejecting erroneous matching pair, obtains optimal matching points and transformation model；
S6: image subject to registration is converted by transformation model, obtains registration result.
2. a kind of weak texture image method for registering based on space time series model according to claim 1, feature
It is, specific step is as follows by step S2:
S21: each pixel on reference picture is selected one by one；
S22: taking image block from reference picture around selected pixel, establishes space temporal model, solving model parameter；
S23: model built applicability is examined using AIC criterion；
S24: based on types of models parameter and detection applicability as a result, describing son to selected pixel feature points.
3. a kind of weak texture image method for registering based on space time series model according to claim 1, feature
It is, step S include thes steps that carrying out conspicuousness detection S25 to characteristic point, and specific step is as follows by step S25:
Reference picture or image subject to registration are divided into several image blocks；
Successively calculate the spy of other each pixels in feature point description and current block of selecting pixel in each image block
The Euclidean distance of sign point description, sums after Euclidean distance value is weighted, obtains weighted sum；
When weighted sum is greater than threshold value, determines that selected pixel is the characteristic point of conspicuousness, remain for image registration, instead
It, selected pixel is not the characteristic point of conspicuousness, abandons it, is not used in image registration.
4. a kind of weak texture image method for registering based on space time series model according to claim 1, feature
It is, specific step is as follows by step S3:
Gradient-norm and the direction for calculating selected characteristic point, with each pixel in the selected feature vertex neighborhood of statistics with histogram
Gradient direction and amplitude, histogram peak direction indicates the principal direction of characteristic point, based on the spy constructed in principal direction and step S2
Sign point description, construction compound characteristics point description.
5. a kind of weak texture image method for registering based on space time series model according to claim 1, feature
It is, in step s 4, signature search is carried out using KD tree or quick nearest neighbor algorithm or K- nearest neighbor algorithm, is found with reference to figure
Corresponding relationship between picture and image characteristic point subject to registration.
6. a kind of weak texture image method for registering based on space time series model according to claim 1, feature
It is, in step s 5, purification processes is carried out to the characteristic point obtained in S4 using random sampling unification algorism, reject mistake
Pairing.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811517008.XA CN109785370B (en) | 2018-12-12 | 2018-12-12 | Weak texture image registration method based on space time sequence model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811517008.XA CN109785370B (en) | 2018-12-12 | 2018-12-12 | Weak texture image registration method based on space time sequence model |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109785370A true CN109785370A (en) | 2019-05-21 |
| CN109785370B CN109785370B (en) | 2023-09-15 |
Family
ID=66496142
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811517008.XA Active CN109785370B (en) | 2018-12-12 | 2018-12-12 | Weak texture image registration method based on space time sequence model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109785370B (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110287972A (en) * | 2019-06-13 | 2019-09-27 | 南京航空航天大学 | A kind of animal painting contours extract and matching process |
| CN112288784A (en) * | 2020-10-09 | 2021-01-29 | 武汉大学 | Descriptor neighborhood self-adaptive weak texture remote sensing image registration method |
| CN112308887A (en) * | 2020-09-30 | 2021-02-02 | 西北工业大学 | Real-time registration method for multi-source image sequence |
| CN112927272A (en) * | 2021-03-30 | 2021-06-08 | 南京工程学院 | Image fusion method based on space general autoregressive model |
| CN117437523A (en) * | 2023-12-21 | 2024-01-23 | 西安电子科技大学 | Weak trace detection method combining SAR CCD and global information capture |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102740082A (en) * | 2011-04-01 | 2012-10-17 | 佳能株式会社 | Image processing apparatus and control method thereof |
| CN103020945A (en) * | 2011-09-21 | 2013-04-03 | 中国科学院电子学研究所 | Remote sensing image registration method of multi-source sensor |
| CN106878687A (en) * | 2017-04-12 | 2017-06-20 | 吉林大学 | A kind of vehicle environment identifying system and omni-directional visual module based on multisensor |
-
2018
- 2018-12-12 CN CN201811517008.XA patent/CN109785370B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102740082A (en) * | 2011-04-01 | 2012-10-17 | 佳能株式会社 | Image processing apparatus and control method thereof |
| CN103020945A (en) * | 2011-09-21 | 2013-04-03 | 中国科学院电子学研究所 | Remote sensing image registration method of multi-source sensor |
| CN106878687A (en) * | 2017-04-12 | 2017-06-20 | 吉林大学 | A kind of vehicle environment identifying system and omni-directional visual module based on multisensor |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110287972A (en) * | 2019-06-13 | 2019-09-27 | 南京航空航天大学 | A kind of animal painting contours extract and matching process |
| CN112308887A (en) * | 2020-09-30 | 2021-02-02 | 西北工业大学 | Real-time registration method for multi-source image sequence |
| CN112308887B (en) * | 2020-09-30 | 2024-03-22 | 西北工业大学 | Multi-source image sequence real-time registration method |
| CN112288784A (en) * | 2020-10-09 | 2021-01-29 | 武汉大学 | Descriptor neighborhood self-adaptive weak texture remote sensing image registration method |
| CN112288784B (en) * | 2020-10-09 | 2022-03-04 | 武汉大学 | Descriptor neighborhood self-adaptive weak texture remote sensing image registration method |
| CN112927272A (en) * | 2021-03-30 | 2021-06-08 | 南京工程学院 | Image fusion method based on space general autoregressive model |
| CN112927272B (en) * | 2021-03-30 | 2023-12-12 | 南京工程学院 | Image fusion method based on space general autoregressive model |
| CN117437523A (en) * | 2023-12-21 | 2024-01-23 | 西安电子科技大学 | Weak trace detection method combining SAR CCD and global information capture |
| CN117437523B (en) * | 2023-12-21 | 2024-03-19 | 西安电子科技大学 | Weak trace detection method combining SAR CCD and global information capture |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109785370B (en) | 2023-09-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109785370A (en) | A kind of weak texture image method for registering based on space time series model | |
| CN107063228B (en) | Target attitude calculation method based on binocular vision | |
| Armangué et al. | Overall view regarding fundamental matrix estimation | |
| JP6216508B2 (en) | Method for recognition and pose determination of 3D objects in 3D scenes | |
| CN109903313B (en) | Real-time pose tracking method based on target three-dimensional model | |
| CN106960449B (en) | Heterogeneous registration method based on multi-feature constraint | |
| JP5385105B2 (en) | Image search method and system | |
| CN104123554B (en) | SIFT image characteristic extracting methods based on MMTD | |
| JP2016197287A (en) | Information processing apparatus, information processing method, and program | |
| CN107492107B (en) | Object identification and reconstruction method based on plane and space information fusion | |
| CN111145232A (en) | Three-dimensional point cloud automatic registration method based on characteristic information change degree | |
| CN110223355B (en) | Feature mark point matching method based on dual epipolar constraint | |
| Belhaoua et al. | Error evaluation in a stereovision-based 3D reconstruction system | |
| CN112017223A (en) | Heterologous image registration method based on improved SIFT-Delaunay | |
| WO2015035462A1 (en) | Point feature based 2d-3d registration | |
| CN112734816B (en) | Heterologous image registration method based on CSS-Delaunay | |
| CN110487254B (en) | Rapid underwater target size measuring method for ROV | |
| Ziqiang et al. | Research of the algorithm calculating the length of bridge crack based on stereo vision | |
| Wan et al. | A performance comparison of feature detectors for planetary rover mapping and localization | |
| Jian-dong et al. | 3D curve structure reconstruction from a sparse set of unordered images | |
| Hui et al. | Surface measurement based on instantaneous random illumination | |
| CN112529960A (en) | Target object positioning method and device, processor and electronic device | |
| Kanuki et al. | Automatic compensation of radial distortion by minimizing entropy of histogram of oriented gradients | |
| CN114877826B (en) | Binocular stereo matching three-dimensional measurement method, system and storage medium | |
| Carrasco et al. | Automated visual inspection using trifocal analysis in an uncalibrated sequence of images |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |