CN109697692A - One kind being based on the similar feature matching method of partial structurtes - Google Patents
One kind being based on the similar feature matching method of partial structurtes Download PDFInfo
- Publication number
- CN109697692A CN109697692A CN201811634213.4A CN201811634213A CN109697692A CN 109697692 A CN109697692 A CN 109697692A CN 201811634213 A CN201811634213 A CN 201811634213A CN 109697692 A CN109697692 A CN 109697692A
- Authority
- CN
- China
- Prior art keywords
- matching
- point
- characteristic point
- neighborhood
- image
- 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
- 238000000034 method Methods 0.000 title claims abstract description 62
- 239000011159 matrix material Substances 0.000 claims abstract description 118
- PXFBZOLANLWPMH-UHFFFAOYSA-N 16-Epiaffinine Natural products C1C(C2=CC=CC=C2N2)=C2C(=O)CC2C(=CC)CN(C)C1C2CO PXFBZOLANLWPMH-UHFFFAOYSA-N 0.000 claims abstract description 96
- 230000008569 process Effects 0.000 claims abstract description 23
- 238000000605 extraction Methods 0.000 claims abstract description 22
- 239000013598 vector Substances 0.000 claims description 36
- 230000014509 gene expression Effects 0.000 claims description 25
- 238000012545 processing Methods 0.000 claims description 13
- 230000009466 transformation Effects 0.000 claims description 11
- 239000012141 concentrate Substances 0.000 claims description 10
- 238000004422 calculation algorithm Methods 0.000 claims description 6
- 238000005259 measurement Methods 0.000 claims description 6
- HUTDUHSNJYTCAR-UHFFFAOYSA-N ancymidol Chemical compound C1=CC(OC)=CC=C1C(O)(C=1C=NC=NC=1)C1CC1 HUTDUHSNJYTCAR-UHFFFAOYSA-N 0.000 claims description 3
- 230000003252 repetitive effect Effects 0.000 claims description 3
- 239000000203 mixture Substances 0.000 claims description 2
- 238000005457 optimization Methods 0.000 abstract description 5
- 238000005516 engineering process Methods 0.000 description 8
- 230000000694 effects Effects 0.000 description 7
- 230000006870 function Effects 0.000 description 6
- 238000002474 experimental method Methods 0.000 description 4
- 238000011160 research Methods 0.000 description 4
- 238000012360 testing method Methods 0.000 description 4
- 230000008859 change Effects 0.000 description 3
- 238000011161 development Methods 0.000 description 2
- 238000005286 illumination Methods 0.000 description 2
- 230000000007 visual effect Effects 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000012217 deletion Methods 0.000 description 1
- 230000037430 deletion Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 238000013508 migration Methods 0.000 description 1
- 230000005012 migration Effects 0.000 description 1
- 238000005295 random walk Methods 0.000 description 1
- 238000000844 transformation Methods 0.000 description 1
- 238000013519 translation Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformations in the plane of the image
- G06T3/14—Transformations for image registration, e.g. adjusting or mapping for alignment of images
- G06T3/147—Transformations for image registration, e.g. adjusting or mapping for alignment of images using affine transformations
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Analysis (AREA)
Abstract
The present invention provides one kind based on the similar feature matching method of partial structurtes, causes matching result undesirable due to noise jamming for solve the problems, such as to occur in process of image registration.Step includes: step 1, carries out feature extraction and initial matching to two image to be matched；Step 2, feature neighborhood of a point affine coefficients matrix is established；Step 3, each of initial matching collection is matched, calculates the difference of feature neighborhood of a point affine coefficients matrix associated therewith；Step 4, neighborhood affine coefficients matrix is optimized, obtains partial structurtes difference degree；Step 5, according to the partial structurtes difference value of each associated characteristic point of matching, setting compares threshold value, determines final characteristic matching to the matching relationship result as image to be matched.The present invention technically overcomes prior art optimization process complexity and restrains slow problem, effectively improves matched efficiency.
Description
Technical field
At image when the invention belongs to match inaccurate under the influence of field of image processing more particularly to picture noise
Reason, it is specially a kind of to be based on the similar feature matching method of partial structurtes.
Background technique
With the fast development of multimedia technology, image has become the important carrier of transmitting information, Digital Image Processing skill
Art increasingly shows its critical role, and wherein image matching technology is even more key content concerned by people in recent years.Cause
For digital image processing techniques other research directions such as image recognition, image retrieval and target identification, target following etc. all
It is further to develop on the basis of image matching technology.It may be said that the progress of image matching technology is able to drive digital picture
The development of processing technique entirety.But image matching technology is not only research hotspot, while being also Research Challenges.Matched target is
Accurate corresponding relationship is found to object identical in image, however its realization process receives many limitations.Such as it is to be matched
Image may be from different photographic equipments, different photographed scenes, or even be different the shooting epoch.Store equipment not
Together, visual angle, the transformation of illumination and background it is mixed and disorderly brought by be object distortion, geometry deformation, noise jamming etc., these
Influence undoubtedly brings huge test to image matching technology.
In recent years, numerous scholars have carried out a large amount of research to image matching technology, and obtain preferable academic
Achievement.General image matching method is carried out based on image characteristic point and feature description vectors.By calculating feature description
The distance between obtain initial matching collection.Since common feature extraction algorithm has good distinction and Scale invariant
Property, it is thus possible to guarantee that it is greatly correct matching pair that resulting initial matching concentration is described by feature.Subsequent calculation
Method is to remove the erroneous matching that initial matching is concentrated by setting geometrical constraint or relation constraint, obtains final correct matching
Collection.The optimization process of set of matches is most commonly realized using figure matching algorithm.Vertex representative image characteristic point in figure, in figure
Side representative image characteristic point association, setting energy function formula indicate point and point, while while between similitude, pass through minimum
Functional expression reaches matching constraint effect.Double optimization is carried out on this basis, selects 3 neighbours' point linear lists around key point
Show the point, obtained coefficient matrix is acted on into the corresponding match point of point, and bring energy function formula into and ask using linear programming
Solution, determines final matching results.But the selection of this method neighbours' point is inflexible, and subsequent solution procedure is excessively complicated, the time
Cost is big, cannot achieve matched efficient.
Summary of the invention
The above matching process there are aiming at the problem that, the present invention provides one kind be based on the similar feature of local linear structure
Matching process.Compared with prior art, the method flexible utilization image structure information, and calculating is simplified during optimization
Amount, substantially increases matched accuracy and recall rate.
Goal of the invention: the problem of being existing matching process Shortcomings to be solved by this invention, proposes a kind of based on office
The similar feature matching method of portion's structure.
Technical solution: the present invention is a kind of to be based on the similar feature matching method of partial structurtes, the method is characterized in that mentioning
It takes characteristics of image description to obtain initial matching collection, coefficient matrix is obtained with key point abutment points linear expression key point, based on being
Matrix number measurement matching deletes confidence using consistency as confidence level judgment criteria to key point regional area Geometrical consistency
Spend small matching pair, it is big to retain confidence level, determines final set of matches.Specifically includes the following steps:
Step 1, feature extraction and initial matching are carried out to two image to be matched, to obtain initial matching corresponding relationship
Collection；
Step 2, identified characteristic point is concentrated to the initial matching obtained in step 1, determines the neighbours of each characteristic point
Point, and feature neighborhood of a point affine coefficients matrix is established accordingly；
Step 3, each of initial matching collection is matched, calculates feature neighborhood of a point affine coefficients associated therewith
The difference of matrix indicates the partial structurtes similitude between each associated characteristic point of matching with this difference.Difference value is smaller,
Partial structurtes similitude is higher, and difference is bigger, and partial structurtes similitude is lower；
Step 4, neighborhood affine coefficients matrix is optimized, definition calculates part by variable of neighborhood affine coefficients matrix
The functional expression of architectural difference degree, the formula of finding a function corresponding neighborhood affine coefficients matrix when taking extreme value, and by this coefficient matrix band
Enter functional expression and calculates characteristic point partial structurtes difference degree；
Step 5, the partial structurtes difference value of the associated characteristic point of each matching obtained according to step 4 sets one
Compare threshold value, retain the matching that difference value is lower than threshold value, deletes the matching that difference value is not less than threshold value, determine final feature
Match the matching relationship result as image to be matched.
Furtherly, detailed step of the invention is as follows:
Step 1, image characteristics extraction and initial matching: the local feature region of image to be matched is extracted, that is, is found in image
To image transformation have higher robustness and detect repetitive rate it is high distinguish key point.To the characteristic point regional area detected
It carries out gradient statistics to calculate, completion describes process to the feature of characteristic point.For any one characteristic point, its descriptor is calculated
Euclidean distance value between the description of other characteristic points selects the corresponding the smallest characteristic point of Euclidean distance value as its matching
Point.Gather the matched point of all characteristic points and forms image initial set of matches；
Step 2, it establishes matching centering key neighborhood of a point affine coefficients matrix: obtaining the base of initial matching collection in step 1
On plinth, to each local feature region in set of matches, find apart from other a certain range of characteristic point conducts of local feature region
Its neighborhood characteristics point, and with these neighborhood characteristics point linear expressions key point, obtain corresponding affine coefficients matrix；
Step 3, the partial structurtes difference degree measurement based on neighborhood affine coefficients matrix: initial to what is obtained in step 1
The matching of each of set of matches obtains feature neighborhood of a point affine coefficients matrix associated with each matching by step 2,
The difference value of two neighborhood affine coefficients matrixes is calculated, what this difference value indicated is the part of each associated characteristic point of matching
Architectural difference degree.Difference value is smaller, and partial structurtes similitude is higher, and difference is bigger, and partial structurtes similitude is lower；
Step 4, neighborhood affine coefficients matrix optimizing: definition is poor as variable calculating partial structurtes using neighborhood affine coefficients matrix
The functional expression of off course degree, the functional expression are made of the sum of two data item, and two data item are respectively initial matching concentration
Difference with the affine combination that corresponding characteristic point and its neighborhood affine coefficients matrix form, the target of solution formula is that function is allowed to take
Value obtains functional expression corresponding neighborhood affine coefficients matrix at extreme point close to zero, and by the affine system of this neighborhood
Matrix number brings functional expression into and calculates characteristic point partial structurtes difference degree；
Step 5, threshold value comparison is carried out to the corresponding partial structurtes difference value of matching each in step 4, when difference value is less than
When defined threshold, corresponding characteristic point to belong to correctly matching pair, when difference value be not less than defined threshold when, corresponding feature
Point concentrates removal erroneous matching pair to erroneous matching pair is belonged to, from initial matching corresponding relationship, and remaining all matchings are to as most
Correct matching result eventually, and export.
Furtherly, the realization process of step 1 is specific as follows: step 1.1, image characteristics extraction: using classical scale
Invariant features transformation (SIFT) feature describes operator extraction characteristics of image.It regard a pair of of image to be matched I and I ' as SIFT algorithm
Input carry out feature extraction, obtain SIFT feature and the corresponding 128 dimensional feature description vectors of characteristic point；
Step 1.2, feature initial matching: the feature vector obtained according to step 1.1, enable two images I to be matched and
The single feature description vectors of I ' are respectively Xi、Xj, the Euclidean distance for calculating two vectors is d (Xi,Xj), and if only if XiAnd other
The distance and d (X of all feature description vectorsi,Xj) ratio be greater than setting threshold value when, it is believed that XiWith XjIt is possible
Match, and ratio be less than or equal to setting threshold value when, then character pair point is considered unmatched.Between feature description vectors
Range formula are as follows:
What wherein k was indicated is 128 dimensional feature description vectors dimension serial numbers, and range is [1,128].XikIndicate vector Xi?
K ties up component, XjkIndicate vector XjKth tie up component；
In other words, this step are as follows: give an image to be matched pair, by feature all in wherein piece image describe to
Amount carries out distance value measurement with the feature description vectors in another piece image one by one, corresponding characteristic point when selecting distance value minimum
As its initial matching, matching each in this way is to the characteristic area that will be corresponded in image to be matched.By all initial matchings to collection
Altogether, the initial matching collection of image to be matched is constituted.
Step 2 is specific as follows: step 2.1, choosing crucial neighborhood point: obtaining image initial set of matches according to step 1.2, enable
M character pair point is arbitrarily matched in set of matches to (p, p '), wherein characteristic point p belongs to image to be matched I, and p ' it is corresponding be to
The point to match in matching image I ' with characteristic point p；Other matchings of matching M adjacent domain are found, and enable corresponding feature
Point is to for (qi,qi')；Then as characteristic point p and qi, while characteristic point p ' and qi' space length value be less than defined threshold when, I
Think characteristic point qiBelong to neighbours' point of characteristic point p, characteristic point p ' belongs to characteristic point qi' neighbours' point, and when its distance value it is big
When being equal to defined threshold, it is believed that characteristic point p and qi, p ' and qi' do not have neighborhood；Neighborhood judgment formula is such as
Under:
||p-qi| | < τ, and | | p'-qi' | | < τ,
Wherein, τ indicates that key point is adjacent similarity threshold a little.The setting of τ is used to ensure image to be matched key
Point p averagely has k arest neighbors, and the setting of k value can be specifically arranged according to the local neighborhood information of key point, so as to more acurrate
The structural information for representing key point.I indicates that the number of neighbours' point of characteristic point p, value range are [1, k], and in testing
The size of k can flexibly be changed by adjusting the value of τ.By carrying out neighbours pass to each characteristic point and other characteristic points
It is that judgment formula calculates, will acquire neighbours' point set of each characteristic point；The neighbour of image to be matched point has been determined by this step
It occupies a little；
Step 2.2, establish matching centering key neighborhood of a point affine coefficients matrix: on the basis of step 2.1, we are really
Neighbours' point of any matching M corresponding characteristic point p in image I is integrated and is expressed as by the neighbours' point for having determined image to be matched pointWherein k indicates the number of neighbours' point of characteristic point p, qiIndicate i-th of neighbours' point of p；It follows and is locally linear embedding into
(Locally Linear Embedding) is theoretical, and the geometry near image characteristic point p can use its neighborhood affine coefficients
Matrix w=[w1,...,wi,...]TIt portrays, then characteristic point p its fixed neighbours' point setAn affine combination
It indicates are as follows: p=∑ wiqi, wherein ∑ wi=1；And for characteristic point p on image I ' corresponding matching characteristic point p ', then use p '
Neighborhood affine coefficients matrix w'=[w'1,...,w'i,...]TIt portrays, so obtain p'=∑ w'iq'i, wherein ∑ w'i=
1, this completes the foundation of crucial vertex neighborhood affine coefficients matrix；
Step 3 is specific as follows: the judgement of match point partial structurtes similitude: the processing by step 2 to image characteristic point obtains
Any associated characteristic point of matching M for having taken initial matching to concentrate neighborhood affine transformation matrix w and w ' respective to (p, p '),
Because of the neighborhood affine coefficients matrix description partial structurtes geometric properties of characteristic point, the neighbour coefficient of matching characteristic point pair
Matrix unanimously indicate is exactly corresponding matching to being correct matching pair, i.e. when w=w ', retain corresponding matching to conduct
Correct matching pair, if the two is unequal, is identified as erroneous matching pair, it is deleted；
Step 4 is specific as follows, optimizes to feature vertex neighborhood affine coefficients matrix: imitative in order to optimize feature neighborhood of a point
Coefficient matrix is penetrated, an optimal neighborhood affine coefficients matrix is foundWith neighbours' point set of characteristic point pLinear combination, then subtract each other with characteristic point p, the difference value both obtained simultaneously carry out normal form processing, then with optimal affine
Coefficient matrix coefficient vectorWith neighbours' point set of the match point p ' of characteristic point pLinear combination,
Subtract each other later with characteristic point p ' and carries out normal form processing；It defines matrix association error J to be made of the sum of two above normal form, is used to table
Show the partial structurtes difference degree of matching double points；Matrix assists the concrete form of error J as follows:
In image to be matched map same target it is correct match in same piece image associated characteristic point in space
It is usually neighbouring on position, and shares similar topology knot due to physical constraint and between two images to be matched
The solution target of structure, matrix association error corresponding in this way is exactly to keep the value of J as small as possible.And corresponding is mistake
It is difficult still to be consistent on different images with the geometry of the Based on Feature Points of its adjacent domain with corresponding characteristic point, because
This, it is judging characteristic point partial structurtes similitude that matrix, which assists error J, i.e. initial matching concentrates matched accuracy to provide one kind
Good method of discrimination.
Step 5 is specific as follows: the difference journey of the partial structurtes of the corresponding characteristic point of each matching based on step 4 acquisition
The value of degree, i.e. matrix association error.A lesser threshold value is set, and provides that the matrix association error when characteristic point is less than the threshold of setting
When value, it is believed that matching associated by this characteristic point is correctly to match, and remains into final set of matches, otherwise, when characteristic point
When matrix covariance is greater than the threshold value of setting, it is believed that matching associated by this characteristic point is the matching of mistake, it is deleted；Finally
By all matching set remained, as final images match result set.
Beneficial technical effect
Feature matching method provided by the present invention is for solving to occur in process of image registration due to noise jamming
And the problem for causing matching result undesirable.Using based on the similar feature matching method of partial structurtes.Including following step
It is rapid: step 1, feature extraction and initial matching to be carried out to two image to be matched, to obtain initial matching set of correspondences；Step
Rapid 2, identified characteristic point is concentrated to the initial matching of acquisition, determines neighbours' point of each characteristic point, and establish feature accordingly
Neighborhood of a point affine coefficients matrix；Step 3, each of initial matching collection is matched, calculates characteristic point associated therewith
Neighborhood affine coefficients matrix difference, indicate that the partial structurtes between the associated characteristic point of each matching are similar with this difference
Property；Step 4, neighborhood affine coefficients matrix is optimized, definition calculates partial structurtes by variable of neighborhood affine coefficients matrix
The functional expression of difference degree takes extreme value to functional expression and further obtains partial structurtes difference degree；Step 5, according to each
Threshold value is compared in partial structurtes difference value with associated characteristic point, setting, retains the matching that difference value is lower than threshold value, and it is poor to delete
Different value is not less than the matching of threshold value, determines final characteristic matching to the matching relationship result as image to be matched.
The present invention is to be based on the similar image matching method of partial structurtes, unavoidable for current image matching technology
Noise jamming problem designs effective solution method.It is defined using the affine relation between image characteristic point and its neighbours' pointIt is special The partial structurtes description of sign point, matches by comparing the partial structurtes deletion error of matching characteristic point, reaches removal noise with this Purpose.The present invention, which primarily rests on, maps the matching of same object to being generally in this think of of adjacent domain in image to be matched On the basis of thinking, noise spot can not meet neighborhood point range and similar partial structurtes two constraints simultaneously, thus can be effective Ground excludes noise.On the other hand, during this present invention specific implementation,The structural information of flexible utilization image itself, gives
Images match is accurately realized in a kind of constraint of affine relation out, and simplifies calculating process, and it is excellent technically to overcome the prior art
Change process is complicated and restrains slow problem, effectively improves matched efficiency.
Detailed description of the invention
Fig. 1 is the basic flow chart of the method for the present invention
Fig. 2 be characterized a little with its adjoint point relation table diagram
Fig. 3 is experiment matching effect figure specific implementation method
The following further describes the present invention with reference to the drawings:
Referring to Fig. 1, one kind being based on the similar matching process of partial structurtes, by successively locating a pair of of image to be matched as follows
Reason:
Step 1, feature extraction and initial matching are carried out to two image to be matched, to obtain initial matching corresponding relationship
Collection；Step 2, identified characteristic point is concentrated to the initial matching obtained in step 1, determines neighbours' point of each characteristic point, and
Feature neighborhood of a point affine coefficients matrix is established accordingly；Step 3, each of initial matching collection is matched, calculates phase therewith
The difference of associated feature neighborhood of a point affine coefficients matrix is indicated between each associated characteristic point of matching with this difference
Partial structurtes similitude.Difference value is smaller, and partial structurtes similitude is higher, and difference is bigger, and partial structurtes similitude is lower；Step
4, neighborhood affine coefficients matrix is optimized, definition calculates partial structurtes difference degree by variable of neighborhood affine coefficients matrix
Functional expression, the formula of finding a function corresponding neighborhood affine coefficients matrix when taking extreme value, and bring this coefficient matrix into functional expression and calculate
Characteristic point partial structurtes difference degree；Step 5, the partial structurtes of the associated characteristic point of each matching obtained according to step 4
Difference value sets a comparison threshold value, retains the matching that difference value is lower than threshold value, deletes the matching that difference value is not less than threshold value,
Determine final characteristic matching to the matching relationship result as image to be matched.
Furtherly, specific steps are as follows: step 1, image characteristics extraction and initial matching: extract the part of image to be matched
Characteristic point, that is, find in image to image transformation have higher robustness and detect repetitive rate it is high distinguish key point.To inspection
The characteristic point regional area measured carries out gradient statistics and calculates, and completion describes process to the feature of characteristic point.For any one
Characteristic point calculates the Euclidean distance value between its descriptor and the description of other characteristic points, selects corresponding Euclidean distance value most
Match point of the small characteristic point as it.Gather the matched point of all characteristic points and forms image initial set of matches；
Step 2, it establishes matching centering key neighborhood of a point affine coefficients matrix: obtaining the base of initial matching collection in step 1
On plinth, to each local feature region in set of matches, find apart from other a certain range of characteristic point conducts of local feature region
Its neighborhood characteristics point, and with these neighborhood characteristics point linear expressions key point, obtain corresponding affine coefficients matrix；
Step 3, the partial structurtes difference degree measurement based on neighborhood affine coefficients matrix: initial to what is obtained in step 1
The matching of each of set of matches obtains feature neighborhood of a point affine coefficients matrix associated with each matching by step 2,
The difference value of two neighborhood affine coefficients matrixes is calculated, what this difference value indicated is the part of each associated characteristic point of matching
Architectural difference degree.Difference value is smaller, and partial structurtes similitude is higher, and difference is bigger, and partial structurtes similitude is lower；
Step 4, neighborhood affine coefficients matrix optimizing: definition is poor as variable calculating partial structurtes using neighborhood affine coefficients matrix
The functional expression of off course degree, functional expression are made of the sum of two data item, and two data item are respectively that initial matching concentrates matching
The difference of the affine combination of corresponding characteristic point and its field affine coefficients matrix composition, the target of solution formula are to allow function value
Close to zero, functional expression corresponding neighborhood affine coefficients matrix at extreme point is obtained, and by this neighborhood affine coefficients
Matrix brings functional expression into and calculates characteristic point partial structurtes difference degree；
Step 5, threshold value comparison is carried out to the corresponding partial structurtes difference value of matching each in step 4, when difference value is less than
When defined threshold, corresponding characteristic point to belong to correctly matching pair, when difference value be not less than defined threshold when, corresponding feature
Point concentrates removal erroneous matching pair to erroneous matching pair is belonged to, from initial matching corresponding relationship, and remaining all matchings are to as most
Correct matching result eventually, and export.
Flow chart as shown in Figure 1, this method are the matching process of a string type: extraction image characteristic point and spy first
Descriptor is levied, image initial set of matches is obtained according to the similitude of description.The neighbor point of each characteristic point is then looked for, and is established
Initial matching concentrates the crucial neighborhood of a point affine coefficients matrix of matching pair, and defines matrix association's error formula to the neighbour of characteristic point
Domain affine coefficients matrix optimizes, and finally retains the matching that homography association error is less than defined threshold, deletes not less than rule
The matching for determining threshold value determines final characteristic matching collection as image to be matched matching result.
Specifically, as shown in Figure 1, the invention discloses one kind based on the similar matching process of local cable architecture.Mainly
The following steps are included:
Step 1, image characteristics extraction and initial matching: the characteristic point and its feature description of image to be matched are extracted, is based on
The similitude of description obtains image initial set of matches.
Step 1.1, image characteristics extraction: using classical rulerDegree invariant features transformation (SIFT) feature describes operator extraction Characteristics of image.By a pair of of image to be matched I and I ' as SIFT algorithm input carry out feature extraction, obtain SIFT feature with And the corresponding 128 dimensional feature description vectors of characteristic point；
Step 1.2, feature initial matching: the feature vector obtained according to step 1.1, enable two images I to be matched and
The single feature description vectors of I ' are respectively Xi、Xj, the Euclidean distance for calculating two vectors is d (Xi,Xj), and if only if XiAnd other
The distance and d (X of all feature description vectorsi,Xj) ratio be greater than setting threshold value when, it is believed that XiWith XjIt is possible
Match, and ratio be less than or equal to setting threshold value when, then character pair point is considered unmatched.Between feature description vectors
Range formula are as follows:
Wherein XikIndicate vector XiKth tie up component, threshold value is taken as 1.1.
Give an image to be matched pair, by feature description vectors all in wherein piece image one by one with another width figure
Feature description vectors as in carry out distance value measurement, when selecting distance value minimum corresponding characteristic point asMatch point.In this way
Each matching is to each characteristic area that will be corresponded in image to be matched.By all initial matchings to gathering, constitute to
Initial matching collection with image.
Step 2, it establishes matching centering key neighborhood of a point affine coefficients matrix: being further illustrated in conjunction with Fig. 2, step 2.1,
It chooses crucial neighborhood point: image initial set of matches being obtained according to step 1.2, enables and arbitrarily matches M character pair point pair in set of matches
(p, p '), wherein characteristic point p belongs to image to be matched I, and it is to match in image to be matched I ' with characteristic point p that p ' is corresponding
Point；Other matchings of matching M adjacent domain are found, and enable corresponding characteristic point to for (qi,qi')；Then as characteristic point p and qi,
Characteristic point p ' and q simultaneouslyi' space length value be less than defined threshold when, it is believed that characteristic point qiBelong to the neighbours of characteristic point p
Point, characteristic point p ' belong to characteristic point qi' neighbours' point, and when its distance value be more than or equal to defined threshold when, it is believed that feature
Point p and qi, p ' and qi' do not have neighborhood；Neighborhood judgment formula is as follows:
||p-qi| | < τ, and | | p'-qi' | | < τ,
Wherein, τ indicates that key point is adjacent similarity threshold a little, value 10.The setting of τ is used to ensure to be matched
Image key points p averagely has k arest neighbors, and the setting of k value can be specifically arranged according to the local neighborhood information of key point, with
Just the structural information of key point is more accurately represented.I indicates that characteristic point p obtains the number of neighbours' point, and value range is [1, k],
And can flexibly change the size of k in testing by adjusting the value of τ, this allows experiment that better effect can be presented.
Step 2.2, establish matching centering key neighborhood of a point affine coefficients matrix: on the basis of step 2.1, we are really
Neighbours' point of any matching M corresponding characteristic point p in image I is integrated and is expressed as by the adjacent office point for having determined image to be matched pointWherein k indicates the number of neighbours' point of characteristic point p, and qi indicates i-th of neighbours' point of p；It follows and is locally linear embedding into
(Locally Linear Embedding) is theoretical, and the geometry near image characteristic point p can use its neighborhood affine coefficients
Matrix w=[w1,...,wi,...]TIt portrays, then characteristic point p its fixed neighbours' point setAn affine combination
It indicates are as follows: p=∑ wiqi, wherein ∑ wi=1；And for characteristic point p on image I ' corresponding matching characteristic point p ', then use p '
Neighborhood affine coefficients matrix w'=[w'1,...,w'i,...]TIt portrays, so obtain p'=∑ w'iq'i, wherein ∑ w'i=
1, this completes the foundation of crucial vertex neighborhood affine coefficients matrix；
Step 3, match point partial structurtes similitude judges: the processing by step 2 to image characteristic point, obtains initial
Any associated characteristic point of matching M in set of matches neighborhood affine transformation matrix w and w ' respective to (p, p '), because of neighborhood
The affine coefficients matrix description partial structurtes geometric properties of characteristic point, so the one of the neighbour coefficient matrix of matching characteristic point pair
Cause to indicate is exactly corresponding matching to being correct matching pair, i.e. when w=w ', retains corresponding matching to as correct matching
It is right, if the two is unequal, it is identified as erroneous matching pair, it is deleted；
Step 4, feature vertex neighborhood affine coefficients matrix is optimized: in order to optimize feature neighborhood of a point affine coefficients square
Battle array finds an optimal neighborhood affine coefficients matrixWith neighbours' point set of characteristic point pLinearly
Combination, then subtracts each other with characteristic point p, and the difference value both obtained simultaneously carries out normal form processing, then with optimal affine coefficients matrix coefficient
VectorWith neighbours' point set of the match point p ' of characteristic point pLinear combination, later and characteristic point
P ' subtracts each other and carries out normal form processing；It defines matrix association error J to be made of the sum of two above normal form, for indicating matching double points
Partial structurtes difference degree；Matrix assists the concrete form of error J as follows:
What the size of J value indicated is the difference degree of the partial structurtes of characteristic point p and p ', and the value of J is smaller to mean that feature
Point p and p' partial structurtes are more similar.The correct matching that same target is mapped from principle, in image to be matched is intended to gather
Collection shares similar topological structure in adjacent domain, and due to physical constraint between images, thus corresponding J value very little.
And the geometric layout of the match point near erroneous matching is difficult still to be consistent on different images, such corresponding J value
With regard to larger.Therefore, matrix association error J provides a kind of good method to assess matched correctness.Setting one is lesser
Compare threshold value, when matrix association error J is less than the threshold value of setting, it is believed that corresponding matching is correctly to match, in initial matching
It concentrates and retains the matching, when threshold value of the matrix association error J not less than setting, it is believed that corresponding matching is the matching of mistake,
Initial matching, which is concentrated, deletes the matching.
The validity of formula J is established in neighborhood affine coefficients matrixLocal geometric variation image is remained unchanged
On the basis of assuming that.In the document of characteristic matching, often assume that corresponding regional area is affine constant.Specific method of proof
As follows: any matching M Corresponding matching point concentrated to initial matching is to (p, p '), and there are its proximity matching MiCorresponding matching point
To (q, q ').Characteristic point p ' can be by its matching characteristic point p by one 2 × 2 rotation scaled matrix A and one 2 × 1 flat
The amount of shifting to t is approached, meanwhile, the characteristic point of adjacent domain has similar transformation, so neighbours' point q of the p of characteristic point can also
To be approached by its match point q ' by rotation scaled matrix A and translation vector t.Then:
WhereinThis constraint ensures the invariance of conversion.Neighborhood affine coefficients matrix is just demonstrated in this wayBetween image local geometric variation remain unchanged, i.e., ifThenIt sets up.If more serious
Image fault appears near key point p or p ', then above-mentioned formula may be invalid.Fortunately, image serious distortion is not
It is common situation.In short, we prove herein, as matching M and its proximity matching MiIt is correctly to match, then regional area structure
It is affine constant.
Since initial matching rally includes erroneous matching, combined structure has respective neighborhood linear in each image
Combined matching characteristic point is to can reduce error caused by noise.And for pure erroneous matching, can also carry out excellent
Change, is handled to provide stronger robustness to the matching under noise jamming.Detailed process is as follows for optimization:
Formula J can be written as:
Wherein X=[p-q1,...,p-qi...], Y=[p'-q'1,...,p'-q'i,...].Set C=X is setTX+
YTY.Then Lagrange multiplier λ is introduced to executeThen formula can convert again are as follows:
Wherein 1=[1 ..., 1]TIt is | N | × 1 column vector.By taking the gradient of J and being set to zero,Value
It may be calculated
FormulaSolution need the inverse of explicit algorithm Matrix C, according to definition, Matrix C is symmetrical positive semi-definite.However, because
Be typically greater than 2 to match the quantity of adjoining set of matches of M, thus Matrix C can be it is unusual.In order to make formula linearly may be used
Solution, in practical applications, a small product of identity matrix I is further added in Matrix C by we as regularization term.
Cnew=C+ ε I,
Compared with the mark of C, ε is arranged to a lesser value, is set as 10 in entire experiment-3tr(C).We can be with
By solving system of linear equationsAnd adjustmentTo obtainTo makeIt is handlingAfterwards, may be used
With what will be calculatedIt brings into the formula of matrix association error J and calculates corresponding J value, and then show that each matching is corresponding special
The partial structurtes difference value of sign point pair.
Step 5, the difference degree of the partial structurtes of the corresponding characteristic point of each matching obtained based on step 4, i.e. matrix
Assist the value of error.A lesser threshold value is set, threshold value value is 8, and provides to be less than setting when the matrix association error of characteristic point
Threshold value when, it is believed that matching associated by this characteristic point is correctly to match, and remains into final set of matches, otherwise, works as feature
When the matrix covariance of point is greater than the threshold value of setting, it is believed that matching associated by this characteristic point makes the matching of mistake, it is deleted；
Finally by all matching set remained, as final images match result set.
Embodiment
Experimental Hardware environment of the invention is: Intel (R) Core (TM) [email protected] 3.30GHz, 8G
Memory, Microsoft Windows7 Ultimate, programmed environment are Visual Studio 2015, MATLAB (R2016a) 64
Position, test chart (being detailed in experiment matching effect figure shown in Fig. 3) derive from South Korea Seoul university (Seoul National
University, SNU) online disclosed multiple target object matching criteria image set.
Select all image of SNU image set, share 6 groups of image to be matched, image name be respectively Books,
Bulletins, Jigsaws, Mickeys, Minnies and Toys include several objects in each image, and often treat
There are different view transformations, different illumination and different Self-variations between matching object, these all make using this
The matching technique of a little images faces very big challenge.
The present invention is on the basis of matching technique is realized and other matching process have carried out pair of accuracy and recall rate
Than the matching process using comparison includes Hough ballot characteristic matching method HV (Feature Matching with Alternate
Hough and Inverted Hough Transforms), discrete topology searches matching method DTS (Discrete Tabu
Search For Graph Matchting), Weighted random migration figure matching method RRWM (Reweighted Random Walks
For Graph Matching) and weak geometrical relationship under spatial match method EWGR (Spatial Matching as
ensemble of weak geometric relations).It is all to describe operator with SIFT and carry out feature to retouch in matching process
It states, and is based on obtaining initial matching collection, to remove the erroneous matching of initial matching concentration as target.In control methods
Parameters select be all corresponding parameter when experiment effect behaves oneself best.A method is showed using accuracy and recall rate
Performance.Comparing result is as follows:
| HV | RRWM | DTS | EWGR | Ours | |
| Precision (%) | 71.36 | 77.68 | 78.22 | 85.19 | 95.45 |
| Recall (%) | 94.04 | 90.05 | 94.17 | 71.49 | 97.50 |
It can be seen that on the Data Representation of accuracy and recall rate from the data in table, matching process of the invention takes
The effect obtained shows well than other matching process, thus demonstrates meaning of the present invention.
Claims (7)
1. one kind is based on the similar feature matching method of partial structurtes, it is characterised in that: by computer by a pair of figure to be matched
As carrying out matching treatment, obtaining has the Feature Points Matching of similar partial structurtes to collection, comprising the following steps:
Step 1, feature extraction and initial matching are carried out to two image to be matched, obtains initial matching set of correspondences；Into
When row feature extraction, the characteristic point of initial matching set of correspondences is obtained；
Step 2, identified characteristic point is concentrated to the initial matching corresponding relationship obtained in step 1, determines each characteristic point
Neighbours' point, and feature neighborhood of a point affine coefficients matrix is established accordingly；
Step 3, each of initial matching collection is matched, calculates feature neighborhood of a point affine coefficients matrix associated therewith
Difference, indicate the partial structurtes similitude between the associated characteristic point of each matching with this difference: difference value is smaller, part
Structural similarity is higher, and difference is bigger, and partial structurtes similitude is lower；
Step 4, neighborhood affine coefficients matrix is optimized, calculates partial structurtes difference by variable of neighborhood affine coefficients matrix
The functional expression of degree, the formula of finding a function corresponding neighborhood affine coefficients matrix when taking extreme value, and bring this coefficient matrix into functional expression
Calculate characteristic point partial structurtes difference degree；
Step 5, the partial structurtes difference value of the associated characteristic point of each matching obtained according to step 4, sets a comparison
Threshold value retains the matching that difference value is lower than threshold value, deletes the matching that difference value is not less than threshold value, determines final characteristic matching pair
Matching relationship result as image to be matched.
2. as described in claim 1 a kind of based on the similar feature matching method of partial structurtes, it is characterised in that: specific steps
Successively are as follows:
Step 1, image characteristics extraction and initial matching: extracting the local feature region of image to be matched, that is, finds in image to figure
As transformation have higher robustness and detect repetitive rate it is high distinguish key point；The characteristic point regional area detected is carried out
Gradient statistics calculates, and completion describes process to the feature of characteristic point；For any one characteristic point, calculate it descriptor and its
Euclidean distance value between the description of his characteristic point selects the corresponding the smallest characteristic point of Euclidean distance value as its match point；
Gather the matched point of all characteristic points and forms image initial set of matches；
Step 2, matching centering key neighborhood of a point affine coefficients matrix is established: on the basis of step 1 obtains initial matching collection,
To each local feature region in set of matches, the neighbour apart from other a certain range of characteristic points of local feature region as it is found
Characteristic of field point, and with these neighborhood characteristics point linear expressions key point, obtain corresponding affine coefficients matrix；
Step 3, the partial structurtes difference degree measurement based on neighborhood affine coefficients matrix: to the initial matching obtained in step 1
The matching of each of collection obtains feature neighborhood of a point affine coefficients matrix associated with each matching by step 2, calculates
The difference value of two neighborhood affine coefficients matrixes, what this difference value indicated is the partial structurtes of each associated characteristic point of matching
Difference degree；Difference value is smaller, and partial structurtes similitude is higher, and difference is bigger, and partial structurtes similitude is lower；
Step 4, neighborhood affine coefficients matrix optimizing: definition calculates partial structurtes difference journey by variable of neighborhood affine coefficients matrix
The functional expression of degree, the functional expression are made of the sum of two data item, and two data item are respectively that initial matching concentrates matching pair
The difference of the affine combination of the characteristic point answered and its neighborhood affine coefficients matrix composition, the solution target of formula are to make function value most
Possibly close to zero, functional expression corresponding neighborhood affine coefficients matrix at extreme point is obtained, and by this neighborhood affine coefficients square
Battle array brings functional expression into and calculates characteristic point partial structurtes difference degree；
Step 5, threshold value comparison is carried out to the corresponding partial structurtes difference value of matching each in step 4:
When difference value be less than defined threshold when, corresponding characteristic point to belong to correctly matching pair；
When difference value is not less than defined threshold, corresponding characteristic point is to erroneous matching pair is belonged to, from initial matching corresponding relationship
Concentrate removal erroneous matching pair, remaining all matchings are exported to as final correct matching result.
3. as claimed in claim 2 a kind of based on the similar feature matching method of partial structurtes, it is characterised in that: step 1 tool
Body is as follows:
Step 1.1, operator extraction image to be matched image characteristics extraction: is described using Scale invariant features transform (SIFT) feature
Feature；A pair of of image to be matched I and I ' is subjected to feature extraction as the input of SIFT algorithm, obtains SIFT feature and spy
The corresponding feature description vectors of sign point；The feature description vectors have 128 dimensions；
Step 1.2, initial characteristics match: the feature description vectors obtained according to step 1.1 enable to be matched in step 1.1
The single feature description vectors of two images I and I ' are respectively Xi、Xj, calculate XiVector and XjThe Euclidean distance of vector is d (Xi,
Xj), and if only if XiWith the distance and d (X of other all feature description vectors in image I 'i,Xj) ratio be greater than setting
When threshold value, X is determinediWith XjCorresponding characteristic point is matched；And ratio is when being less than or equal to the threshold value of setting, then character pair point
It is considered unmatched；The distance between feature description vectors formula are as follows:
What wherein k was indicated is 128 dimensional feature description vectors dimension serial numbers, and range is [1,128]；XikIndicate vector XiKth dimension
Component, XjkIndicate vector XjKth tie up component；
By the feature extraction to image to be matched I and I ' in step 1.1, and in step 1.2 feature description between away from
It is calculated from value, the initial images match relationship of final output is to set.
4. as claimed in claim 2 a kind of based on the similar feature matching method of partial structurtes, it is characterised in that: step 2 tool
Body is as follows:
Step 2.1, it chooses crucial neighborhood point: image initial set of matches being obtained according to step 1.2, enables and arbitrarily matches M in set of matches
Character pair point is to (p, p '), and wherein characteristic point p belongs to image to be matched I, and p ' it is corresponding be in image to be matched I ' with spy
The point that sign point p matches；Other matchings of matching M adjacent domain are found, and enable corresponding characteristic point to for (qi,qi')；Then
As characteristic point p and qi, while characteristic point p ' and qi' space length value be less than defined threshold when, it is believed that characteristic point qiBelong to
Neighbours' point of characteristic point p, characteristic point p ' belong to characteristic point qi' neighbours' point, and when its distance value be more than or equal to defined threshold when,
It is considered that characteristic point p and qi, p ' and qi' do not have neighborhood；Neighborhood judgment formula is as follows:
||p-qi| | < τ, and | | p'-qi' | | < τ,
Wherein, τ indicates that key point is adjacent similarity threshold a little；The setting of τ is used to ensure that image to be matched key point p is flat
K arest neighbors is all had, the setting of k value can be specifically arranged according to the local neighborhood information of key point, so as to more accurate table
The structural information of key point is shown；I indicates that the number of neighbours' point of characteristic point p, value range are [1, k]；By to each spy
Sign point carries out the calculating of neighborhood judgment formula with other characteristic points, will acquire the adjacent Ju Dianji of each characteristic point；Pass through this
Step has determined the adjacent office point of image to be matched characteristic point；Step 2.2, matching centering key neighborhood of a point affine coefficients square is established
Battle array: on the basis of step 2.1, it is determined that the adjacent office point of image to be matched characteristic point, any matching M is corresponding in image I
Neighbours' point of characteristic point p integrated be expressed asWherein k indicates the number of neighbours' point of characteristic point p, and qi indicates p's
I-th of neighbours' point；It follows and is locally linear embedding into (Locally Linear Embedding) theory, near image characteristic point p
Geometry can use its neighborhood affine coefficients matrix w=[w1,...,wi,...]TIt portrays, then characteristic point p is had determined that with it
Neighbours' point setOne it is affine combination be expressed as: p=∑ wiqi, wherein ∑ wi=1；And for characteristic point p in image
Corresponding matching characteristic point p ' on I ' then uses the neighborhood affine coefficients matrix w'=[w' of p '1,...,w'i,...]TIt portrays, institute
To obtain p'=∑ w'iq'i, wherein ∑ w'i=1；The foundation of crucial vertex neighborhood affine coefficients matrix is completed in this way.
5. one kind as described in claim 2 or 4 is any is based on the similar feature matching method of partial structurtes, it is characterised in that:
Step 3 is specific as follows:
The judgement of match point partial structurtes similitude: the processing by step 2 to image characteristic point obtains initial matching concentration
Any matching associated characteristic point of M neighborhood affine transformation matrix w and w ' respective to (p, p '), because of neighborhood affine coefficients square
Battle array describes the partial structurtes geometric properties of characteristic point, so the neighbour coefficient matrix of matching characteristic point pair unanimously indicates just
It is corresponding matching to being correct matching pair, is i.e. when w=w ', retains corresponding matching to as correct matching pair, if two
Person is unequal, then is identified as erroneous matching pair, it is deleted.
6. one kind as described in claim 2 or 4 is any is based on the similar feature matching method of partial structurtes, it is characterised in that:
Step 4 is specific as follows:
Feature vertex neighborhood affine coefficients matrix is optimized: in order to optimize feature neighborhood of a point affine coefficients matrix, finding one
A optimal neighborhood affine coefficients matrixWith neighbours' point set of characteristic point pLinear combination, with spy
Sign point p subtracts each other, and obtains the difference value of the two and carries out normal form processing, then with optimal affine coefficients matrix coefficient vectorWith neighbours' point set of the match point p ' of characteristic point pLinear combination, later with characteristic point p ' phase
Subtract and carries out normal form processing；It defines matrix association error J to be made of the sum of two above normal form, for indicating the part of matching double points
Architectural difference degree；Matrix assists the concrete form of error J as follows:
7. as claimed in claim 2 a kind of based on the similar feature matching method of partial structurtes, it is characterised in that: step 5 tool
Body is as follows:
Obtaining initial matching according to step 4 concentrates the corresponding matrix of each characteristic point to assist error amount, sets a threshold value, and advise
It is fixed:
When the matrix of characteristic point association error is less than the threshold value of the setting, it is believed that matching associated by this characteristic point is correct
Match, and remains into final set of matches；
Otherwise, when the matrix covariance of characteristic point is greater than the threshold value of setting, it is believed that matching associated by this characteristic point makes mistake
Matching, it is deleted；
Finally by all matching set remained, as final images match result set and export.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811634213.4A CN109697692B (en) | 2018-12-29 | 2018-12-29 | Feature matching method based on local structure similarity |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811634213.4A CN109697692B (en) | 2018-12-29 | 2018-12-29 | Feature matching method based on local structure similarity |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109697692A true CN109697692A (en) | 2019-04-30 |
| CN109697692B CN109697692B (en) | 2022-11-22 |
Family
ID=66233012
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811634213.4A Active CN109697692B (en) | 2018-12-29 | 2018-12-29 | Feature matching method based on local structure similarity |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109697692B (en) |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110245671A (en) * | 2019-06-17 | 2019-09-17 | 艾瑞迈迪科技石家庄有限公司 | A kind of endoscopic images characteristic point matching method and system |
| CN110472543A (en) * | 2019-08-05 | 2019-11-19 | 电子科技大学 | One kind is based on the matched mechanical drawing control methods of local connection features |
| CN110659654A (en) * | 2019-09-24 | 2020-01-07 | 福州大学 | Drawing duplicate checking and plagiarism preventing method based on computer vision |
| CN110807797A (en) * | 2019-10-22 | 2020-02-18 | 中国测绘科学研究院 | Multi-source heterogeneous surface entity and point entity matching method considering global optimization and storage medium thereof |
| CN110874849A (en) * | 2019-11-08 | 2020-03-10 | 安徽大学 | Non-rigid point set registration method based on local transformation consistency |
| CN111461196A (en) * | 2020-03-27 | 2020-07-28 | 上海大学 | Method and device for identifying and tracking fast robust image based on structural features |
| CN112348105A (en) * | 2020-11-17 | 2021-02-09 | 贵州省环境工程评估中心 | Unmanned aerial vehicle image matching optimization method |
| CN114399422A (en) * | 2021-12-02 | 2022-04-26 | 西安电子科技大学 | Remote sensing image registration method based on local information and global information |
| CN115049847A (en) * | 2022-06-21 | 2022-09-13 | 上海大学 | Characteristic point local neighborhood characteristic matching method based on ORB descriptor |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20050238198A1 (en) * | 2004-04-27 | 2005-10-27 | Microsoft Corporation | Multi-image feature matching using multi-scale oriented patches |
| CN103839253A (en) * | 2013-11-21 | 2014-06-04 | 苏州盛景空间信息技术有限公司 | Arbitrary point matching method based on partial affine transformation |
| CN105354578A (en) * | 2015-10-27 | 2016-02-24 | 安徽大学 | Multi-target object image matching method |
-
2018
- 2018-12-29 CN CN201811634213.4A patent/CN109697692B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20050238198A1 (en) * | 2004-04-27 | 2005-10-27 | Microsoft Corporation | Multi-image feature matching using multi-scale oriented patches |
| CN103839253A (en) * | 2013-11-21 | 2014-06-04 | 苏州盛景空间信息技术有限公司 | Arbitrary point matching method based on partial affine transformation |
| CN105354578A (en) * | 2015-10-27 | 2016-02-24 | 安徽大学 | Multi-target object image matching method |
Non-Patent Citations (1)
| Title |
|---|
| 鲍文霞等: "结合亮度序局部特征描述的图匹配算法", 《哈尔滨工程大学学报》 * |
Cited By (17)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110245671B (en) * | 2019-06-17 | 2021-05-28 | 艾瑞迈迪科技石家庄有限公司 | Endoscope image feature point matching method and system |
| CN110245671A (en) * | 2019-06-17 | 2019-09-17 | 艾瑞迈迪科技石家庄有限公司 | A kind of endoscopic images characteristic point matching method and system |
| CN110472543A (en) * | 2019-08-05 | 2019-11-19 | 电子科技大学 | One kind is based on the matched mechanical drawing control methods of local connection features |
| CN110472543B (en) * | 2019-08-05 | 2022-10-25 | 电子科技大学 | Mechanical drawing comparison method based on local connection feature matching |
| CN110659654A (en) * | 2019-09-24 | 2020-01-07 | 福州大学 | Drawing duplicate checking and plagiarism preventing method based on computer vision |
| CN110807797A (en) * | 2019-10-22 | 2020-02-18 | 中国测绘科学研究院 | Multi-source heterogeneous surface entity and point entity matching method considering global optimization and storage medium thereof |
| CN110807797B (en) * | 2019-10-22 | 2022-03-22 | 中国测绘科学研究院 | Multi-source heterogeneous surface entity and point entity matching method considering global optimization and storage medium thereof |
| CN110874849A (en) * | 2019-11-08 | 2020-03-10 | 安徽大学 | Non-rigid point set registration method based on local transformation consistency |
| CN110874849B (en) * | 2019-11-08 | 2023-04-18 | 安徽大学 | Non-rigid point set registration method based on local transformation consistency |
| CN111461196A (en) * | 2020-03-27 | 2020-07-28 | 上海大学 | Method and device for identifying and tracking fast robust image based on structural features |
| CN111461196B (en) * | 2020-03-27 | 2023-07-21 | 上海大学 | Rapid robust image identification tracking method and device based on structural features |
| CN112348105A (en) * | 2020-11-17 | 2021-02-09 | 贵州省环境工程评估中心 | Unmanned aerial vehicle image matching optimization method |
| CN112348105B (en) * | 2020-11-17 | 2023-09-01 | 贵州省环境工程评估中心 | Unmanned aerial vehicle image matching optimization method |
| CN114399422A (en) * | 2021-12-02 | 2022-04-26 | 西安电子科技大学 | Remote sensing image registration method based on local information and global information |
| CN114399422B (en) * | 2021-12-02 | 2023-08-08 | 西安电子科技大学 | Remote sensing image registration method based on local information and global information |
| CN115049847A (en) * | 2022-06-21 | 2022-09-13 | 上海大学 | Characteristic point local neighborhood characteristic matching method based on ORB descriptor |
| CN115049847B (en) * | 2022-06-21 | 2024-04-16 | 上海大学 | ORB descriptor-based feature point local neighborhood feature matching method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109697692B (en) | 2022-11-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109697692A (en) | One kind being based on the similar feature matching method of partial structurtes | |
| CN105844669B (en) | A kind of video object method for real time tracking based on local Hash feature | |
| Kim et al. | Boundary preserving dense local regions | |
| CN111242221B (en) | Image matching method, system and storage medium based on image matching | |
| CN106682700B (en) | Block rapid matching method based on key point description operator | |
| CN107358259B (en) | Covering ice for insulator detection method based on GLOH description son and GVF-Snake model | |
| Liu et al. | Improved RANSAC features image‐matching method based on SURF | |
| CN104573681A (en) | Face recognition method | |
| CN103345760B (en) | A kind of automatic generation method of medical image object shapes template mark point | |
| Fu et al. | Novel affine‐invariant curve descriptor for curve matching and occluded object recognition | |
| Guo et al. | Exploring GIS knowledge to improve building extraction and change detection from VHR imagery in urban areas | |
| CN105913069B (en) | A kind of image-recognizing method | |
| Bai et al. | An efficient indexing scheme based on k-plet representation for fingerprint database | |
| Cui et al. | A novel robust feature descriptor for multi-source remote sensing image registration | |
| CN109840529A (en) | A kind of image matching method based on local sensitivity confidence level estimation | |
| Li et al. | Fast SIFT algorithm based on Sobel edge detector | |
| CN108764245A (en) | A method of it improving brand logo similarity and judges accuracy | |
| Li et al. | A genetic-optimized multi-angle normalized cross correlation SIFT for automatic remote sensing registration | |
| CN109117841A (en) | Scene text detection method based on stroke width transformation and convolutional neural networks | |
| Ahmad et al. | A fusion of labeled-grid shape descriptors with weighted ranking algorithm for shapes recognition | |
| Chen et al. | Geometric and non-linear radiometric distortion robust multimodal image matching via exploiting deep feature maps | |
| Liu et al. | An image retrieval algorithm based on multiple convolutional features of RPN and weighted cosine similarity | |
| Wu et al. | Image Matching Algorithm Based on Topology Consistency of Bidirectional Optimal Matching Point Pairs. | |
| Guo et al. | Image registration method based on improved SIFT algorithm and essential matrix estimation | |
| Shi et al. | Feature matching using sequential evaluation on sample consensus |
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 |