CN106886794B - Take the heterologous remote sensing image homotopy mapping method of high-order structures feature into account - Google Patents
Take the heterologous remote sensing image homotopy mapping method of high-order structures feature into account Download PDFInfo
- Publication number
- CN106886794B CN106886794B CN201710078970.7A CN201710078970A CN106886794B CN 106886794 B CN106886794 B CN 106886794B CN 201710078970 A CN201710078970 A CN 201710078970A CN 106886794 B CN106886794 B CN 106886794B
- Authority
- CN
- China
- Prior art keywords
- point
- image
- matching
- characteristic point
- feature
- 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.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/74—Image or video pattern matching; Proximity measures in feature spaces
- G06V10/75—Organisation of the matching processes, e.g. simultaneous or sequential comparisons of image or video features; Coarse-fine approaches, e.g. multi-scale approaches; using context analysis; Selection of dictionaries
- G06V10/757—Matching configurations of points or features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/13—Satellite images
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Multimedia (AREA)
- Health & Medical Sciences (AREA)
- Remote Sensing (AREA)
- Astronomy & Astrophysics (AREA)
- Artificial Intelligence (AREA)
- Computing Systems (AREA)
- Databases & Information Systems (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Software Systems (AREA)
- Image Analysis (AREA)
Abstract
The invention belongs to Surveying Science and Technology fields, provide a kind of heterologous remote sensing image homotopy mapping method for taking high-order structures feature into account, including the data preparation before the heterologous Image Matching of progress；Carry out the division and extraction of characteristic point cluster；The determination of characteristic point candidate matches point；Take the homotopy mapping of high-order structures feature into account；To every layer of pyramid image matching result, the RFM model area net adjusted data method of fusion iteration method with variable weights is utilized；Matching result successively refine until completing raw video layer, finally realizes the automatic reliable matching of heterologous remote sensing image same place.Geometrical constraint feature and high-order structures feature are introduced into layer-by-layer pyramid image matching, both ensure that the sparsity of hypergraph model by integrated use rational function model and hypergraph Image Matching model of the present invention, are also minimized super side sampling bring information loss；The present invention improves the reliability and success rate of heterologous Remote Sensing Images Matching, and the workload that same place manually measures is effectively reduced.
Description
Technical field
The invention belongs to Surveying Science and Technology fields, are related to heterologous remote sensing image homotopy mapping method, more particularly to
A kind of heterologous remote sensing image homotopy mapping method for taking high-order structures feature into account.
Background technique
With the rapid development of earth observation technology, a plurality of types of sensors occur like the mushrooms after rain, they are with more
High temporal resolution, spatial resolution record the variation of earth surface, but also the image of covering areal is increasingly
Multiplicity.The target alignment by union of heterologous remote sensing image has become one of the important research direction in Photogrammetry and Remote Sensing field.So
And either single-point three-dimensional localization or block adjustment, it is required to quickly and accurately look for image same place, and measure it
Image coordinate.
For image matching method, related scholar is had conducted extensive research, and is broadly divided into three kinds: feature-based matching,
Matching based on region and the matching based on overall image.Feature-based matching method can theoretically reduce image ash
The adverse effect of difference is spent, there is better adaptability.In recent years, related scholar proposes, as wavelet character, Shape context,
Gradient radial direction angle pyramid histogram, SIFT (scale invariant feature conversion, Scale-invariant feature
Transform, SIFT), the various features such as SURF (accelerate robust features, Speed Up Robust Features, SURF) retouch
State operator.However, the matching process based on such feature operator, with computationally intensive, successful match rate is not high, match point distribution
The disadvantages of uneven；The matching algorithm based on edge, Block Characteristic that some scholars propose, by image radiation characteristic difference and
The influence of noise, extracts that feature of the same name is very difficult, and universality is not strong, is not able to satisfy remote sensing image aerial triangulation operation
In same place position distribution is uniformly required.
Matching process based on region usually matches the gray scale or gradient information of entire image or image sub-district.This
Class algorithm is that the gray scale to be overlapped in the target area and the field of search that contain respective image on image is basic as matching, utilizes certain
Similarity measure determines the matching degree of two image blocks.However, Image Matching or big portion either based on gray areas
Point the Image Matching based on feature, be all based on the Image Matching of single-point or part, can not be effectively ensured matching result can
By property.
Matching process based on overall image, by between mapping relations or proximity matching intrinsic between consideration matching result
Compatible consistency, come achieve the purpose that improve matching reliability.The it is proposeds such as Chen Sa be based on the domain Contoulet Hausdorff away from
From the multi-source Remote Sensing Images matching process with population, the correspondence between point set is established using least square Hausdorff distance
Relationship enhances matched reliability.Li Yuqian etc. proposes the method for registering images based on level set, by mapping function by SAR
Characteristics of image and optical signature combine, and construct energy functional model, solve curve evolvement equation, Lai Shixian using Level Set Method
Heterologous Image registration.But for remote sensing image, influenced by factors such as image coverage area broadness, hypsographies, same place it
Between relationship be difficult to carry out effective expression with function mapping relations, and due to the spatial sparsity of match point, will cause compatible
It is consistent to assume no longer to set up completely, for this purpose, these whole matching methods are directly introduced into heterologous Remote Sensing Images Matching and are not conformed to
It is suitable.
In terms of Remote Sensing Images Matching Method, the patent No. is that the patent of invention of ZL201010242888.1 is (a kind of steady
Automatic matching method for high-resolution satellite image connecting points) and the patent No. be the patent of invention (base of ZL201110091756.8
In the multi-source satellite-borne SAR image automatic matching method of RFM model) all refer to remote sensing image homotopy mapping.But its research emphasis
It is to be provided during the bundle block adjustment method based on RFM model is dissolved into homotopy mapping using RFM model
Geometrical constraint, improve characteristic point initial point position prediction precision, the constraint of core line geometry is provided reduces matching search space and base
Error matching points are rejected in RFM model area net adjusted data, but it substantially still falls within single-point matching, is not introduced into matching process
With the high-order structures feature between point.
Summary of the invention
In order to solve the above technical problems in background technology, the purpose of the present invention is to provide one kind can be improved
The reliability of heterologous Remote Sensing Images Matching and the heterologous remote sensing image homotopy mapping side for taking high-order structures feature into account of success rate
High-order structures feature provided by RFM model between geometrical constraint and match point is introduced into matching process by method, this method
In, it realizes automatic, reliable, the accurate matching of same place, substantially increases the degree of automation and working efficiency of same place acquisition,
Greatly reduce labor workload.
To achieve the above object, the present invention adopts the following technical scheme:
A kind of heterologous remote sensing image homotopy mapping method for taking high-order structures feature into account, it is characterised in that: described to take into account
The heterologous remote sensing image homotopy mapping method of high-order structures feature the following steps are included:
1) data preparation before heterologous Image Matching: the data before the heterologous Image Matching include generating pyramid shadow
Picture；The pyramid image includes raw video layer and non-primary image bearing layer；The non-primary image bearing layer and raw video
Layer is sequentially overlapped from top to bottom and forms pyramid structure；The raw video layer is the 0th layer of pyramid image；
2) the raw video layer on pyramid image carries out the division and extraction of characteristic point cluster: by the whole of raw video layer
Width raw video is uniformly distributed a certain size sub-district, inside each sub-district, several equally distributed grid of further division；?
Within the scope of each grid, mentioned in each grid using the method for Forstner feature extraction operator point common in photogrammetric
Best features point is taken, characteristic point point cluster is formed；
3) determination of Feature Points Matching candidate point: on every layer of pyramid image carry out the initial point of characteristic point prediction,
Establish core line geometry constraint equation, the geometry of match window image is slightly corrected and based on the matching that normalized mutual information is estimated,
Determine the corresponding matching candidate point of each characteristic point；
4) take the homotopy mapping of high-order structures feature into account: to the corresponding matching candidate point of each characteristic point in step 3)
Determination after the completion of, using characteristic point point cluster as unit on every layer of pyramid image, while considering corresponding to each characteristic point
The grey similarity feature of matching candidate point and the geometrical characteristic of high-order, it is true by the global image matching method based on hypergraph
Determine the corresponding same place of each characteristic point；The grey similarity of matching candidate point corresponding to each characteristic point is characterized in one
Rank feature；The order of the high-order is not less than two ranks；
5) block adjustment based on RFM model is carried out in every layer of pyramid image of non-primary image bearing layer, is rejected wrong
Object coordinates corresponding to matching candidate point accidentally and calculating matching candidate point；
6) matching of pyramid image is successively completed from top to bottom until completing raw video layer, finally realizes heterologous remote sensing
The automatic reliable matching of image same place.
Preferably, the specific implementation of step 4) of the present invention is:
4.1) hypergraph model is constructed；
4.2) super side sampling is carried out to the obtained hypergraph model of step 1)；
4.3) hypergraph model is resolved according to sampled result, determines that each characteristic point is corresponding of the same name by calculation result
Point.
Preferably, the specific implementation of step 4.1) of the present invention is:
If some characteristic point point cluster is P1, feature point number N1, matching candidate point set corresponding to the characteristic point is combined into
P2, feature point number N2,Indicate set P1And P2Ith feature point；Set P1And P2Corresponding hypergraph
Model is G=(V, E, A) and G '=(V ', E ', A ')；
Wherein:
V and V ' is respectively corresponding characteristic point vertex set, i.e.,
For the set on super side, d is that super side includes feature point number；
A and A ' is property set corresponding to super side；
If characteristic point point cluster P1The vertex response sets of corresponding matching candidate point areA k tuple of C
As shown in formula (1):
cs1=(v1,v′1) ..., cs1=(vk,v′k) (1)
Wherein:
(v1,…,vk)∈V；
(v′1,…,v′k)∈V′；
Super side e is enabled respectively1,…,kExpression contains vertex v1,…,vkSide, super side e '1,…,kExpression contains vertex v
′1,…,v′kSide；The similitude of hypergraph matching for a k rank figure, the super side for being k by comparing two degree is first to measure k
The similitude of group；Specified k ties up similarity measurements flow function fk, the k dimension similarity measurements flow function fkParameter be hypergraph attribute set
Attribute vector in A and A '；The super side of more low order can be considered simultaneously, at this time the similarity tensor on the super side of k rank are as follows:
Tsi (1)=f1(ai,a′i) (2)
In formula:
Parameter γ(k)Indicate the weighting coefficient of the super side similarity of k rank；
The upper right mark (k) of T indicates the dimension of tensor；
a1,…,k,a′1,…,kRespectively indicate the vector parameter in k rank hypergraph attribute set；
If super side maximum order is δ, similitude tensor TδFor high order tensor, the similitude on all not unison super sides is contained
Information.
Preferably, the specific implementation of step 4.2) of the present invention is:
To matching candidate point corresponding to characteristic point point cluster and characteristic point, established using k-d tree (k-dimensional tree)
Spatial index is played, to each characteristic point of characteristic point point cluster, by the way of stochastical sampling, randomly selects and a certain number of includes
3 tuples of this feature point surpass side as it；And where Feature Points Matching candidate point on image, and non-sampled all ternarys
Group, but only consider 3 tuples being made of the included characteristic point Corresponding matching candidate point in the super side of this feature point sampling, and 3 tuples
Middle any two point does not repeat.
Preferably, the specific implementation of step 4.3) of the present invention is:
Hypergraph matching, is to find optimal response relation in the response sets C of vertex, with a binary system assignment matrix
Description,The two-way constraint of its general satisfaction, i.e.,To binary system assignment
Matrix X vectorization, is indicated with x, then for x, the corresponding total similarity of Image Matching are as follows:
Score (x) obtains x corresponding to maximum value*Vector corresponding to as optimal binary system assignment matrix；
Formula (4) can use tensor product representation are as follows:
Hypergraph matching is solved using tensor power iteration algorithm, is obtained in characteristic point point cluster corresponding to each characteristic point
Same place.
Preferably, the matched specific implementation side estimated in step 3) of the present invention based on normalized mutual information
Formula is:
On every layer of pyramid image centered on the initial point of characteristic point, a rectangular window is opened up, generates search
Window image；Established core line geometry constraint equation is utilized, each pixel of search window image is traversed, calculates search first
The pixel currently traversed in rope window image to core line equation distance, when the distance of the pixel to core line equation be less than it is given
When threshold value, match window is opened up centered on the pixel；When the geometry deformation of match window is more than specified threshold, to matching window
Image in mouthful carries out geometry and slightly corrects, and the geometry deformation of the match window includes rotation angle deformation parameter and pantograph ratio
Example deformation parameter；Then, estimated using normalized mutual information, it is similar to pixel each in search window to calculate characteristic point to be matched
Degree；To the normalized mutual information measure value of each pixel of search window, local maximum, normalizing are extracted using non-maximum restraining algorithm
Changing mutual information measure local maximum to be greater than pixel corresponding to specified threshold is the point of matching candidate corresponding to characteristic point；When
When the point number of matching candidate corresponding to characteristic point is greater than 5, normalization mutual trust is pressed to matching candidate corresponding to characteristic point
It ceases measure value and carries out size sequence, take first 5；Finally, by the result of the point of matching candidate corresponding to characteristic point according to characteristic point
Point cluster is that unit carries out data organization.
Preferably, the data before heterologous Image Matching provided by the present invention further include calculating heterologous remote sensing as needed
The RPC parameter of image.
Preferably, the acquisition modes of the RPC parameter of heterologous remote sensing image provided by the present invention are:
Using the method unrelated with landform utilize image rigorous geometric model, based on different elevation faces generate it is intensive and
Equally distributed virtual controlling grid, presses the principle of least square using virtual controlling and is resolved, and obtains the RPC ginseng of image
Number.
Compared with prior art, the present invention has following remarkable advantage and effect:
The present invention provides a kind of heterologous remote sensing image homotopy mapping methods for taking high-order structures feature into account, carry out first
Data preparation before heterologous Image Matching, calculate as needed heterologous remote sensing image RPC (rational polynominal coefficient,
Rational Polynomial Coefficient, RPC) parameter and generate pyramid image；Then, characteristic point cluster is carried out
It divides and extracts, and in each pyramid image layer, using the initial point of image RPC parameter prediction point to be matched, establish approximation
Core line geometry, geometry slightly correct match window image, carry out the matching estimated based on normalized mutual information, find the time of characteristic point
Match point is selected, then using characteristic point point cluster as unit, the homotopy mapping for taking high-order structures feature into account is carried out, by based on hypergraph
Global image matching method, with the corresponding optimal same place of each characteristic point of determination.To every layer of pyramid image matching result, benefit
With the RFM model area net adjusted data method of fusion iteration method with variable weights, realize the refining of image RPC parameter, error matching points from
Dynamic rejecting and matching double points answer the calculating of object coordinates.Matching result successively is refined until raw video layer, is finally realized different
The automatic reliable matching of source remote sensing image same place.The present invention is on the basis of using for reference the prior art, to feature point extraction strategy
It improves, marks off several equally distributed sub-districts in image, extract characteristic point point cluster, integrated use rational function in sub-district
Geometrical constraint feature and high-order structures feature are introduced into layer-by-layer pyramid image and matched by model and hypergraph Image Matching model
In, global image matching is carried out to characteristic point point cluster, can be taken into account simultaneously between the grey similarity and match point of match point
High-order structures feature only samples the super side of characteristic point in characteristic point point cluster, for the feature of Image Matching to matching candidate
The super side of point all considers, both ensure that the sparsity of hypergraph model, is also minimized super side sampling bring information loss；This
Match point wrong in matching is effectively deleted in invention, improves the reliability and success rate of heterologous Remote Sensing Images Matching, is effectively dropped
The workload that low same place manually measures.
Detailed description of the invention
Fig. 1 is the process of the heterologous remote sensing image homotopy mapping method provided by the present invention for taking high-order structures feature into account
Schematic diagram.
Specific embodiment
Technical solution provided by the present invention is described in further detail with reference to the accompanying drawing:
Embodiment 1:
Referring to Fig. 1, the present invention provides a kind of heterologous remote sensing image homotopy mapping method for taking high-order structures feature into account,
Each step is elaborated as follows:
Data preparation before the heterologous Image Matching of step 1):
Data preparation before heterologous Image Matching, specifically includes that
1) calculate the RPC parameter of heterologous remote sensing image: this step is optional step, only when remote sensing image only provides it is strictly several
What carried out when model." the High Resolution Remote Sensing Satellites of Zhang Yongsheng, the superfine work of Gong Dan (can refer to using the method unrelated with landform
Using --- imaging model, Processing Algorithm and application technology "), using the rigorous geometric model of image, looked unfamiliar based on different elevations
At intensive and equally distributed virtual controlling grid, press the principle of least square using virtual controlling and resolved, obtain shadow
The RPC parameter of picture.
2) pyramid image generates: using practical, simple 3 × 3 pixel method of average, heterologous remote sensing image is generated 3 grades
Pyramid image, raw video layer are the 0th layer of pyramid image.
The division and extraction of step 2) characteristic point cluster:
In order to ensure image feature point even distribution, and can be used for subsequent taking match point high-order structures feature into account
Global image matching, the present embodiment are first uniformly distributed whole picture raw video in raw video layer 4 × 4 or 5 × 5 certain big
Small sub-district is further divided into 7 × 7 or 10 × 10 equally distributed grid inside each sub-district.In each grid range
It is interior, using Forstner feature extracting method common in photogrammetric (can refer to that Zhang Zuxun, Zhang Jianqing write " number is taken the photograph
Shadow surveying ") a best features point is extracted in each grid.If the characteristic information in certain grid is unobvious, by net
Lattice central point is as characteristic point, to form characteristic point point cluster, each sub-district is an independent characteristic point point cluster unit.Together
When, characteristic point point cluster is stored as document form, is used for subsequent Image Matching.
The determination of step 3) Feature Points Matching candidate point:
The determination of Feature Points Matching candidate point carries out on every layer of pyramid image, mainly includes the initial point of characteristic point
Prediction, establish core line geometry constraint equation, the geometry of match window image is slightly corrected and is estimated based on normalized mutual information
Matching.Wherein, the prediction of the initial point of characteristic point, establish core line geometry constraint equation, the geometry of match window image slightly entangles
Positive concrete methods of realizing, can refer to the patent of invention that the patent No. is ZL201110091756.8, (multi-source based on RFM model is spaceborne
SAR image automatic matching method).
For the matching estimated based on normalized mutual information, it is with the initial point of characteristic point on every layer of pyramid image
A rectangular window is opened up at center, generates search window image；Established core line geometry constraint equation is utilized, to search window
Each pixel traversal of mouthful image, calculate first the pixel to core line equation distance, when the pixel to the distance of core line equation
When less than given threshold value, match window is opened up centered on the pixel.When the geometry deformation of match window is more than specified threshold
When, the geometry for carrying out match window image is slightly corrected, and the geometry deformation of match window includes rotation angle deformation parameter and contracting
Put scale distortion parameter.Then, estimated using normalized mutual information, calculate each pixel in characteristic point to be matched and search window
Similarity.To the normalized mutual information measure value of each pixel of search window, local maximum is extracted using non-maximum restraining algorithm,
It is matching candidate point that normalized mutual information, which estimates local maximum greater than pixel corresponding to specified threshold,.When matching candidate point
When number is greater than 5, normalized mutual information measure value is pressed to matching candidate and carries out size sequence, takes first 5.Finally, by special
The result of sign point matching candidate point is that unit carries out data organization according to characteristic point point cluster.
Step 4) takes the homotopy mapping of high-order structures feature into account:
The homotopy mapping for taking high-order structures feature into account carries out on every layer of pyramid image, Feature Points Matching candidate point
Determination after, carried out one by one as unit of characteristic point point cluster.Specific step is as follows:
1) building of hypergraph model: some characteristic point point cluster is set as P1, feature point number N1, corresponding characteristic point
It is P with candidate point set2, it is N comprising feature point number2,Indicate set P1And P2Ith feature point.Collection
Close P1And P2Corresponding hypergraph model is G=(V, E, A) and G '=(V ', E ', A ').Wherein, V and V ' is corresponding characteristic point top
Point set, i.e., For the set on super side, d is super side
Include feature point number.A and A ' is property set corresponding to super side.If characteristic point point cluster P1Corresponding candidate same place vertex is rung
It should collect and be combined intoShown in a k tuple of C such as formula (1):
cs1=(v1,v′1) ..., cs1=(vk,v′k) (1)
Wherein, (v1,…,vk) ∈ V, (v '1,…,v′k)∈V′.Super side e is enabled respectively1,…,kExpression contains vertex v1,…,
vkSide, super side e '1,…,kExpression contain vertex v '1,…,v′kSide.Hypergraph matching for a k rank figure, can pass through
The similitude for comparing the similitude on the super side that two degree are k to measure k tuple.Specified k ties up similarity measurements flow function fk, its ginseng
Number is the attribute vector in hypergraph attribute set A and A '.The super side of more low order can be considered simultaneously, at this time the phase on the super side of k rank
Like degree tensor are as follows:
Tsi (1)=f1(ai,a′i) (2)
In formula, parameter γ(k)Indicate that the weighting coefficient of the super side similarity of k rank, the upper right mark (k) of T indicate the dimension of tensor.
a1,…,k,a′1,…,kRespectively indicate the vector parameter in k rank hypergraph attribute set.If super side maximum order is δ, similitude tensor Tδ
For high order tensor, the affinity information on all not unison super sides is contained.
In the present embodiment, the characteristics of according to heterologous Remote Sensing Images Matching, it is similar with the super side building hypergraph of 3 ranks only to choose 1 rank
Spend tensor.Similarity measurements flow function used by the super side of 1 rank is estimated for normalized mutual information, that is, considers the gray scale of Feature Points Matching
Similitude.The used super side similarity measurements flow function of 3 rank, the geometric similarity invariance of main considering feature, calculation formula (3)
It is as follows:
Wherein, ε is constant term,The interior angle of triangle corresponding to the super side of respectively 3 ranks.
2) sampling on super side: in the present embodiment, in order to reduce computation complexity, while retaining between node that mainly connection is closed
System guarantees the matched correct matching rate of hypergraph, the specific steps are as follows: to characteristic point point cluster and Feature Points Matching candidate point, uses
K-d tree sets up spatial index, to each characteristic point of characteristic point point cluster, by the way of stochastical sampling, randomly selects certain
3 tuples comprising this feature point of quantity surpass side as it, 50 are taken in the present embodiment, i.e., the super number of edges of each feature point sampling
It is 50, where Feature Points Matching candidate point on image, and non-sampled all triples, but only consider by this feature point
In the super side of sampling, 3 tuples that included characteristic point Corresponding matching candidate point is constituted, and any two point does not weigh in 3 tuples
It is multiple.This method not only improves hypergraph model sparsity, while the characteristics of utilization Image Matching, by all associated super Bian Junna
Enter to consider, avoids constituting Feature Points Matching candidate point information loss brought by super side progress stochastical sampling.
3) resolving of hypergraph model: hypergraph matching, is exactly to find optimal response relation in the response sets C of vertex, can
It is described with a binary system assignment matrix,The two-way constraint of its general satisfaction, i.e., It to binary system assignment matrix X vectorization, is indicated with x, then for x, corresponding image
Match total similarity are as follows:
For this purpose, score (x) obtains x corresponding to maximum value*Vector corresponding to as optimal binary system assignment matrix.
Formula (4) can use tensor product representation are as follows:
For this purpose, hypergraph matching problem can be solved using tensor power iteration algorithm, each spy in characteristic point point cluster is obtained
The corresponding same place of sign point.
Each characteristic point point cluster is handled one by one, completes the matching of current pyramid image layer characteristic point.
Step 5) is based on RFM model area net adjusted data deletion error match point:
The present embodiment carries out satellite image block adjustment using image space affine Transform Model, auxiliary in iteration method with variable weights,
During adjustment, using the weight for reasonably adjusting each observation, guarantee that error matching points do not influence adjustment result, and realize
Its automatic detection and positioning, specific method refer to the patent of invention that the patent No. is ZL201110091756.8 and (are based on RFM model
Multi-source satellite-borne SAR image automatic matching method (SAR, synthetic aperture radar, Synthetic Aperture Radar)).
It is matched using upper layer matching result constraint lower layer's pyramid image, repeats step 3), step 4), step 5), judgement
Whether raw video layer image matching is completed, until output matching result.
Claims (4)
1. a kind of heterologous remote sensing image homotopy mapping method for taking high-order structures feature into account, it is characterised in that: described to take height into account
The heterologous remote sensing image homotopy mapping method of stage structure feature the following steps are included:
1) data preparation before heterologous Image Matching: the data before the heterologous Image Matching include generating pyramid image；Institute
Stating pyramid image includes raw video layer and non-primary image bearing layer；The non-primary image bearing layer and raw video layer are from upper
It is sequentially overlapped under and and forms pyramid structure；The raw video layer is the 0th layer of pyramid image；
2) the raw video layer on pyramid image carries out the division and extraction of characteristic point cluster: the whole picture of raw video layer is former
Beginning image is uniformly distributed a certain size sub-district, inside each sub-district, several equally distributed grid of further division；Each
Within the scope of grid, extracted most in each grid using the method for Forstner feature extraction operator point common in photogrammetric
Good characteristic point forms characteristic point point cluster；
3) determination of Feature Points Matching candidate point: the prediction of the initial point of characteristic point is carried out on every layer of pyramid image, is established
Core line geometry constraint equation, match window image geometry slightly correct and based on the matching that normalized mutual information is estimated, determine
The corresponding matching candidate point of each characteristic point；
4) take the homotopy mapping of high-order structures feature into account: really to the corresponding matching candidate point of each characteristic point in step 3)
After the completion of fixed, using the characteristic point of raw video layer point cluster as unit on every layer of pyramid image, while each characteristic point is considered
The grey similarity feature of corresponding matching candidate point and the geometrical characteristic of high-order, pass through the overall image based on hypergraph
Method of completing the square determines the corresponding same place of each characteristic point；The grey similarity of matching candidate point corresponding to each characteristic point
It is characterized in single order feature；The order of the high-order is not less than two ranks；
5) block adjustment based on RFM model is carried out in every layer of pyramid image, reject mistake matching candidate point and
Calculate object coordinates corresponding to matching candidate point；
6) matching of pyramid image is successively completed from top to bottom until completing raw video layer, finally realizes heterologous remote sensing image
The automatic reliable matching of same place；
The specific implementation of the step 4) is:
4.1) hypergraph model is constructed；
4.2) super side sampling is carried out to the obtained hypergraph model of step 4.1)；
4.3) hypergraph model is resolved according to sampled result, the corresponding same place of each characteristic point is determined by calculation result；
The specific implementation of the step 4.1) is:
If some characteristic point point cluster is P1, feature point number N1, matching candidate point set corresponding to the characteristic point is combined into P2,
Feature point number is N2,Indicate set P1And P2Ith feature point；Set P1And P2Corresponding hypergraph model
For G=(V, E, A) and G '=(V ', E ', A ')；
Wherein:
V and V ' is respectively corresponding characteristic point vertex set, i.e.,
For the set on super side, d is that super side includes feature point number；
A and A ' is property set corresponding to super side；
If the vertex response sets of the corresponding matching candidate point of characteristic point point cluster P1 areA k tuple of C is such as public
Shown in formula (1):
cs1=(v1, v '1) ..., cs1=(vk, v 'k) (1)
Wherein:
(v1..., vk)∈V；
(v′1..., v 'k)∈V′；
Super side e is enabled respectively1 ..., kExpression contains vertex v1..., vkSide, super side e '1 ..., kExpression contain vertex v '1...,
v′kSide；Hypergraph matching for a k rank figure spends the similitude on the super side for being k by comparing two to measure k tuple
Similitude；Specified k ties up similarity measurements flow function fk, the k dimension similarity measurements flow function fkParameter be hypergraph attribute set A and
Attribute vector in A '；The super side of more low order can be considered simultaneously, at this time the similarity tensor on the super side of k rank are as follows:
Tsi (1)=f1(ai, a 'i) (2)
In formula:
Parameter γ(k)Indicate the weighting coefficient of the super side similarity of k rank；
The upper right mark (k) of T indicates the dimension of tensor；
a1 ..., k, a '1 ..., kRespectively indicate the vector parameter in k rank hypergraph attribute set；
If super side maximum order is δ, similitude tensor TδFor high order tensor, the affinity information on all not unison super sides is contained；
The specific implementation of the step 4.2) is:
To matching candidate point corresponding to characteristic point point cluster and characteristic point, sky is set up using k-d tree (k-dimensional tree)
Between index, to each characteristic point of characteristic point point cluster, by the way of stochastical sampling, randomly select a certain number of comprising the spy
3 tuples of sign point surpass side as it；And where Feature Points Matching candidate point on image, and non-sampled all triples, and
It is 3 tuples for only considering to be made of the included characteristic point Corresponding matching candidate point in the super side of this feature point sampling, and any in 3 tuples
Two points do not repeat；
The specific implementation of the step 4.3) is:
Hypergraph matching, is to find optimal response relation in the response sets C of vertex, is described with a binary system assignment matrix,The two-way constraint of its general satisfaction, i.e.,To binary system assignment matrix X
Vectorization is indicated with x, then for x, the corresponding total similarity of Image Matching are as follows:
Score (x) obtains x corresponding to maximum value*Vector corresponding to as optimal binary system assignment matrix；
Formula (4) can use tensor product representation are as follows:
Hypergraph matching is solved using tensor power iteration algorithm, is obtained of the same name corresponding to each characteristic point in characteristic point point cluster
Point.
2. the heterologous remote sensing image homotopy mapping method according to claim 1 for taking high-order structures feature into account, feature
It is: is based on the matched specific implementation that normalized mutual information is estimated in the step 3):
On every layer of pyramid image centered on the initial point of characteristic point, a rectangular window is opened up, generates search window
Image；Established core line geometry constraint equation is utilized, each pixel of search window image is traversed, first calculating search window
The pixel currently traversed in mouthful image arrives the distance of core line equation, when the distance of the pixel to core line equation is less than given threshold value
When, match window is opened up centered on the pixel；When the geometry deformation of match window is more than specified threshold, in match window
Image carry out geometry and slightly correct, the geometry deformation of the match window includes that rotation angle deformation parameter and scaling become
Shape parameter；Then, estimated using normalized mutual information, calculate the similarity of each pixel in characteristic point to be matched and search window；
To the normalized mutual information measure value of each pixel of search window, local maximum, normalization are extracted using non-maximum restraining algorithm
It is the point of matching candidate corresponding to characteristic point that mutual information measure local maximum, which is greater than pixel corresponding to specified threshold,；Work as spy
When the corresponding matching candidate point number of sign point is greater than 5, normalized mutual information is pressed to matching candidate corresponding to characteristic point
Measure value carries out size sequence, takes first 5；Finally, by the result of the point of matching candidate corresponding to characteristic point according to characteristic point point
Cluster is that unit carries out data organization.
3. the heterologous remote sensing image homotopy mapping method according to claim 2 for taking high-order structures feature into account, feature
Be: the data before the heterologous Image Matching further include calculating the RPC parameter of heterologous remote sensing image as needed.
4. the heterologous remote sensing image homotopy mapping method according to claim 3 for taking high-order structures feature into account, feature
Be: the acquisition modes of the RPC parameter of the heterologous remote sensing image are:
The rigorous geometric model that image is utilized using the method unrelated with landform is generated intensive and uniform based on different elevation faces
The virtual controlling grid of distribution, presses the principle of least square using virtual controlling and is resolved, and obtains the RPC parameter of image.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710078970.7A CN106886794B (en) | 2017-02-14 | 2017-02-14 | Take the heterologous remote sensing image homotopy mapping method of high-order structures feature into account |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710078970.7A CN106886794B (en) | 2017-02-14 | 2017-02-14 | Take the heterologous remote sensing image homotopy mapping method of high-order structures feature into account |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106886794A CN106886794A (en) | 2017-06-23 |
| CN106886794B true CN106886794B (en) | 2019-06-14 |
Family
ID=59179621
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710078970.7A Active CN106886794B (en) | 2017-02-14 | 2017-02-14 | Take the heterologous remote sensing image homotopy mapping method of high-order structures feature into account |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106886794B (en) |
Families Citing this family (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109409292B (en) * | 2018-10-26 | 2021-09-03 | 西安电子科技大学 | Heterogeneous image matching method based on refined feature optimization extraction |
| CN109934298B (en) * | 2019-03-19 | 2022-10-28 | 安徽大学 | Progressive graph matching method and device of deformation graph based on clustering |
| CN110555817B (en) * | 2019-09-10 | 2022-06-24 | 中国科学院遥感与数字地球研究所 | Geometric normalization method and device for remote sensing image |
| CN111414988B (en) * | 2020-02-13 | 2023-04-18 | 湖北工业大学 | Remote sensing image super-resolution method based on multi-scale feature self-adaptive fusion network |
| CN111683221B (en) * | 2020-05-21 | 2021-08-03 | 武汉大学 | Real-time video monitoring method and system for natural resources embedded with vector red line data |
| CN114565839A (en) * | 2022-02-17 | 2022-05-31 | 广州市城市规划勘测设计研究院 | Remote sensing image target detection method, device, equipment and computer medium |
| CN114998399B (en) * | 2022-05-20 | 2023-04-11 | 中国人民解放军61540部队 | Heterogeneous optical remote sensing satellite image stereopair preprocessing method |
| CN115471728B (en) * | 2022-10-12 | 2023-04-18 | 湖北工业大学 | Fruit rapid three-dimensional positioning method based on multi-order geometric constraint integral matching |
| CN116521927B (en) * | 2023-06-30 | 2024-02-13 | 成都智遥云图信息技术有限公司 | Remote sensing image matching method and system based on network map tiles |
Family Cites Families (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101915913B (en) * | 2010-07-30 | 2012-09-19 | 中交第二公路勘察设计研究院有限公司 | Steady automatic matching method for high-resolution satellite image connecting points |
| CN102073874B (en) * | 2010-12-29 | 2013-04-24 | 中国资源卫星应用中心 | Geometric constraint-attached spaceflight three-line-array charged coupled device (CCD) camera multi-image stereo matching method |
| CN102213762B (en) * | 2011-04-12 | 2012-11-28 | 中交第二公路勘察设计研究院有限公司 | Method for automatically matching multisource space-borne SAR (Synthetic Aperture Radar) images based on RFM (Rational Function Model) |
| CN103077527A (en) * | 2013-02-05 | 2013-05-01 | 湖北工业大学 | Robust multi-source satellite remote sensing image registration method |
| KR102018046B1 (en) * | 2014-02-24 | 2019-09-04 | 한국전자통신연구원 | Method and apparatus for extracting image feature |
| CN103822616B (en) * | 2014-03-18 | 2016-01-20 | 武汉大学 | A kind of figure segmentation retrains with topographic relief the Remote Sensing Images Matching Method combined |
-
2017
- 2017-02-14 CN CN201710078970.7A patent/CN106886794B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN106886794A (en) | 2017-06-23 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106886794B (en) | Take the heterologous remote sensing image homotopy mapping method of high-order structures feature into account | |
| CN105469388B (en) | Building point cloud registration method based on dimensionality reduction | |
| CN109840553B (en) | Extraction method and system of cultivated land crop type, storage medium and electronic equipment | |
| CN103218783B (en) | Satellite remote sensing images fast geometric correcting method based on control point image database | |
| CN103337052B (en) | Automatic geometric correcting method towards wide cut remote sensing image | |
| CN103115614B (en) | Associated parallel matching method for multi-source multi-track long-strip satellite remote sensing images | |
| CN111047695B (en) | Method for extracting height spatial information and contour line of urban group | |
| CN110033484A (en) | Set high extracting method to the high closed forest sample of a kind of combination UAV image and TLS point cloud | |
| Zhang et al. | Efficient registration of terrestrial LiDAR scans using a coarse-to-fine strategy for forestry applications | |
| CN103235810B (en) | Remote sensing image reference mark data intelligence search method | |
| CN102959946A (en) | Augmenting image data based on related 3d point cloud data | |
| Zhong et al. | A method for extracting trees from vehicle-borne laser scanning data | |
| CN109100719A (en) | Combine plotting method with the topographic map of optical image based on satellite-borne SAR image | |
| CN114241464A (en) | Cross-view image real-time matching geographic positioning method and system based on deep learning | |
| CN113468982B (en) | City function classification method, device and storage medium | |
| Park et al. | Conceptual design of the CZMIL data processing system (DPS): algorithms and software for fusing lidar, hyperspectral data, and digital images | |
| CN107563366A (en) | A kind of localization method and device, electronic equipment | |
| Sun et al. | Building outline extraction from aerial imagery and digital surface model with a frame field learning framework | |
| CN112561981A (en) | Photogrammetry point cloud filtering method fusing image information | |
| Sun et al. | Individual tree segmentation and biomass estimation based on UAV Digital aerial photograph | |
| Wang et al. | Building extraction from LiDAR and aerial images and its accuracy evaluation | |
| Wu et al. | Building Facade Reconstruction Using Crowd-Sourced Photos and Two-Dimensional Maps | |
| Bauchet et al. | Rooftops or footprints? reliable building footprint extraction from high-resolution satellite images | |
| Hwang et al. | 3D modeling and accuracy assessment-a case study of photosynth | |
| CN118640878B (en) | Topography mapping method based on aviation mapping technology |
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 |