CN102800098A - Multi-characteristic multi-level visible light full-color and multi-spectrum high-precision registering method - Google Patents
Multi-characteristic multi-level visible light full-color and multi-spectrum high-precision registering method Download PDFInfo
- Publication number
- CN102800098A CN102800098A CN2012102514871A CN201210251487A CN102800098A CN 102800098 A CN102800098 A CN 102800098A CN 2012102514871 A CN2012102514871 A CN 2012102514871A CN 201210251487 A CN201210251487 A CN 201210251487A CN 102800098 A CN102800098 A CN 102800098A
- Authority
- CN
- China
- Prior art keywords
- image
- full
- sift characteristic
- sift
- multispectral 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 66
- 238000001228 spectrum Methods 0.000 title abstract 6
- 230000009466 transformation Effects 0.000 claims abstract description 82
- 238000000605 extraction Methods 0.000 claims abstract description 13
- 238000000354 decomposition reaction Methods 0.000 claims abstract description 12
- 230000008878 coupling Effects 0.000 claims description 28
- 238000010168 coupling process Methods 0.000 claims description 28
- 238000005859 coupling reaction Methods 0.000 claims description 28
- 238000006243 chemical reaction Methods 0.000 claims description 14
- 239000013598 vector Substances 0.000 claims description 14
- 239000000284 extract Substances 0.000 claims description 12
- 238000005070 sampling Methods 0.000 claims description 12
- 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 4
- 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 description 3
- 238000013461 design Methods 0.000 claims description 2
- 239000004744 fabric Substances 0.000 claims description 2
- 230000001131 transforming effect Effects 0.000 abstract 1
- 230000008569 process Effects 0.000 description 10
- 238000010586 diagram Methods 0.000 description 4
- 230000008901 benefit Effects 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 239000011159 matrix material Substances 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 241001269238 Data Species 0.000 description 1
- 240000001439 Opuntia Species 0.000 description 1
- 230000000903 blocking effect Effects 0.000 description 1
- 238000004040 coloring Methods 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 230000002950 deficient Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 238000007500 overflow downdraw method Methods 0.000 description 1
- 230000000505 pernicious effect Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000001737 promoting effect Effects 0.000 description 1
- 230000003252 repetitive effect Effects 0.000 description 1
- 239000002699 waste material Substances 0.000 description 1
Images
Abstract
The invention discloses a visible light full-color image and multi-spectrum image registering method which comprises the following steps of: performing multi-scale decomposition of the full-color image and multi-spectrum image to generate low-resolution full-color image and multi-spectrum image; extracting and matching SIFT characteristics from the low-resolution images and removing the exterior points; obtaining an initial transformation model by use of the matched SIFT characteristic pair and the iteration re-weighting least square method; performing SIFT characteristic extraction and matching as well as exterior point removal based on image block pairs on the original image by use of the geometric constraint provided by the initial transformation model, and selecting the optimal transformation type from all SIFT characteristic pair sets and obtaining the transformation parameter by use of the iteration re-weighting least square method; and transforming the multi-spectrum image according to the transformation model to obtain the transformed multi-spectrum image.
Description
Technical field
The present invention relates to a kind of panchromatic method for registering, especially space flight, method for registering of remote sensing image that the airborne sensor platform obtains with multispectral image of visible light that be used for.
Background technology
Rely on high spatial resolution full-colour image or low spatial resolution multispectral image all can't obtain comprehensive, the real information of target separately, must give particulars simultaneously information and color characteristic, human eye could accurately be judged the target that comprises in the remote sensing images.Yet,, can not obtain the multispectral image of high spatial resolution simultaneously owing to the restriction of hardware.For this reason, general satellite is all taken high spatial resolution full-colour image and low spatial resolution multispectral image simultaneously.Utilize registration, integration technology; Can be with the multispectral image of spatial resolution full-colour image and low spatial resolution multispectral image generation high spatial resolution; The image that generates had both kept the spatial resolution characteristic of full-colour image; Incorporated the colouring information of multispectral image again, thereby provided convenience for declaring the figure personnel.Yet, compare with the high-definition image that the hand-held digital details is taken, remote sensing images second-rate, the registration difficulty increases.For the large scale remote sensing images, the existence of repetitive structure (like buildings) has brought challenge for traditional feature matching method.Yet the traditional image fusion method has very high requirement (generally requiring registration accuracy to be superior to 0.5 pixel) to image registration accuracy.In this case, develop automatic, practical remote sensing images high registration accuracy system, become particularly urgent and necessary.
Aspect remote sensing image registration, researcher and technician have carried out big quantity research both at home and abroad, and have released some Related products such as Erdas and Envi.These softwares can obtain good registration accuracy on external satellite data such as QuickBird, WorldView.Yet because the singularity of homemade satellite image quality, these business softwares are also bad to the registration performance of homemade satellite.Therefore caused such situation: on the one hand, country strengthens dynamics of investment, greatly develops homemade satellite, and the commercial satellite data are provided free to the demestic user; On the other hand, because homemade satellite image registration difficulty is very big, a lot of users would rather spend to buy external satellite data, and a large amount of homemade satellite datas are by idle.This situation has caused the huge waste of state fund, resource; If this trend continues to continue, also can cause China to external pernicious dependence, all totally unfavorable to national defence, economy, development in science and technology.In fact, even Erdas and Envi software, employing also be traditional method for registering, do not consider the specific (special) requirements of large scale remote sensing image registration.Therefore, for the large scale remote sensing images, the existing software processing speed is all very slow; Even can't accomplish the registration of large-size images because of memory problem.Therefore, must have the high registration accuracy method of independent intellectual property right, satisfy the needs of practical application at aspects such as precision, speed, automaticities to homemade satellite data research.
Summary of the invention
The technical matters that (one) will solve
Technical matters to be solved by this invention is to the singularity of homemade satellite image and the existing business software deficiency to the large-size images registration, provide a kind of automatically, remote sensing images high registration accuracy method efficiently
(2) technical scheme
The present invention proposes the method for registering of a kind of visible light full-colour image and multispectral image, comprises the steps:
Step S1, full-colour image and multispectral image are carried out multiple dimensioned decomposition, generate the full-colour image and the multispectral image of low resolution;
Step S2, said low-resolution image on extract, coupling SIFT characteristic and remove exterior point, the SIFT characteristic of utilizing coupling is to obtaining the initial transformation model with the heavy weighted least-squares method of iteration;
Step S3, original image on utilize said initial transformation model according to the geometrical constraint that provides; Carry out removing, and on all SIFT characteristic pair sets, select optimum alternative types and utilize the heavy weighted least-squares method of iteration to try to achieve transformation parameter based on the right SIFT feature extraction of image block, coupling and exterior point;
Step S4: according to said transformation model multispectral image is carried out conversion, obtain the multispectral image after the conversion.
(3) beneficial effect
The present invention can effectively solve homemade satellite image registration problems through based on the panchromatic and multispectral High Precision Automatic method for registering of other visible light of many feature multi-levels, remedies the defective of external business software to large scale remote sensing image processing aspect.Method for registering of the present invention has good versatility and practicality to homemade satellite image, and the widespread use with promoting homemade satellite greatly has good economic benefits.
Description of drawings
Fig. 1 is a system chart of the present invention;
Fig. 2 is the illustration of extraction SIFT characteristic of a specific embodiment of method of the present invention;
Fig. 3 is that DOG makes up legend in the SIFT feature extraction of a specific embodiment of method of the present invention;
Fig. 4 is that extreme point extracts legend in the SIFT feature extraction of a specific embodiment of method of the present invention;
Fig. 5 is that principal direction detects legend in the SIFT feature extraction of a specific embodiment of method of the present invention;
Fig. 6 is a synoptic diagram for unique point structure description vector of a specific embodiment of the present invention.
Embodiment
For making the object of the invention, technical scheme and advantage clearer,, and, the present invention is done further detailed description with reference to accompanying drawing below in conjunction with specific embodiment.
Other panchromatic high registration accuracy method with multispectral image of many feature multi-levels of the present invention both can realize by hardware mode, also can realize by software mode.For example on personal computer, industrial computer and server, install and carry out, also can method of the present invention make embedded chip and embody with the form of hardware with the form of software.Specific embodiments of the invention describes with reference to the accompanying drawings.
In the following description, described " image " refers in particular to the remote sensing images that obtain through remote sensing equipment, and is to have carried out digitized digital picture.Yet the present invention is not limited to remote sensing images, and for the full-colour image that needs registration and the multispectral image that obtain in other technical field, the present invention is also applicable.
Fig. 1 has provided an embodiment system chart, and is as shown in Figure 1, and generally speaking, method of the present invention comprises the steps:
Step S1: multiple dimensioned decomposition.Full-colour image and multispectral image are carried out multiple dimensioned decomposition, generate the full-colour image and the multispectral image of low resolution;
According to a kind of embodiment of the present invention, the multiple dimensioned decomposition among this embodiment realizes through falling sampling.Though multiple dimensioned decomposition can be by accomplished in many ways such as wavelet pyramid, gaussian pyramids, calculated amount is big more than falling sampling; And in the smart registration stage, SIFT (Scale Invariant Feature Transform) characteristic also will be extracted on original image.Therefore, when keeping registration accuracy, also significantly reduced calculated amount based on the multiple dimensioned is olation that falls sampling.
Step S2: thick registration.Said low-resolution image on extract, coupling SIFT characteristic and remove exterior point, the SIFT characteristic of utilizing coupling is to obtaining the initial transformation model with the heavy weighted least-squares method of iteration;
According to a kind of embodiment of the present invention, adopt the arest neighbors ratioing technigue to mate the SIFT characteristic, and remove exterior point, the SIFT characteristic that exterior point refers to not satisfy above-mentioned transformation model is right;
Transformation model comprises alternative types and transformation parameter two layers of meaning.Equally, the definite of transformation model comprises that also alternative types is selected and transformation parameter is found the solution.According to a kind of embodiment of the present invention, the alternative types in thick registration stage is an affined transformation, utilizes the heavy weighted least-squares method of iteration to obtain transformation parameter.Said transformation model is that the initial value of subsequent registration transformation model is the initial transformation model, all will constantly adjust in the type and the parameter thereof of smart registration phase transformation model.
Step S3: smart registration.Original image on utilize said initial transformation model according to the geometrical constraint that provides; Carry out removing, and on all SIFT characteristic pair sets, select optimum alternative types and utilize the heavy weighted least-squares method of iteration to try to achieve transformation parameter based on the right SIFT feature extraction of image block, coupling and exterior point;
According to a kind of embodiment of the present invention, original image on adopt based on the right SIFT feature extraction of image block, coupling and exterior point and remove.
According to a kind of embodiment of the present invention, the alternative types of smart registration is a kind of in affined transformation and the projective transformation, utilizes the heavy weighted least-squares method of iteration to select alternative types and ask for transformation parameter.
Step S4: image transformation.According to said transformation model multispectral image is carried out conversion, obtain the multispectral image after the conversion.
According to a kind of embodiment of the present invention; According to transformation model multispectral image is carried out the bicubic spline interpolation; Calculate the coordinate of the maximum overlapping region of the multispectral image after reference picture and the conversion then according to the size of transformation model and full-colour image and multispectral image, take out, preserve the interior full-colour image and the multispectral image of coordinate of said maximum overlapping region.
For making the object of the invention, technical scheme and advantage clearer,, the present invention is done further detailed description below in conjunction with specific embodiment.
In this embodiment, multiple dimensioned decomposition realizes according to following flow process:
Step S11: if the known image latitude and longitude information, the overlapping region of taking out full-colour image and multispectral image according to latitude and longitude information; If there is not the image latitude and longitude information, utilize the entire image information of two width of cloth images to carry out registration;
Step S12: the multiple dimensioned decomposition number of plies n that confirms multispectral image according to the size of multispectral image
1, confirm the multiple dimensioned decomposition number of plies n of full-colour image according to the full-colour image size and with the spatial resolution difference of multispectral image
2
Step S13: sampling is fallen in full-colour image and multispectral image, and full-colour image falls the sampling multiple and falls the sampling multiple for
for
multispectral image
Step S2 of the present invention is the registration on low resolution full-colour image and low resolution multispectral image.In this embodiment, thick registration is realized according to following flow process:
S21: feature extraction.On low resolution full-colour image and low resolution multispectral image, extract the SIFT characteristic respectively, obtain two SIFT characteristic set A and B.Each SIFT characteristic is the proper vector of 132 dimensions, the x coordinate that preceding 4 dimensional features are respectively unique points, y coordinate, optimal scale, principal direction, and back 128 dimensional features are normalized SIFT character representation vectors of this unique point;
S22: characteristic matching.Euclidean distance and arest neighbors ratioing technigue based on SIFT feature description vector mate two SIFT characteristic sets: certain the SIFT characteristic a among the pair set A; Euclidean distance between the SIFT character representation vector distance of its arest neighbors SIFT feature b 1 and inferior neighbour SIFT feature b 2 in set B is respectively t1 and t2; If t1/t2＜0.8, then a and b1 are that the SIFT characteristic is right.
S23: exterior point is removed.Full-colour image under the low resolution and the alternative types between the multispectral image are regarded as affined transformation, utilize RANSAC (RANdom SAmple Consensus, random sampling consistance) algorithm to remove exterior point, it is right to obtain SIFT characteristic with a high credibility.
S24: affine transformation parameter is found the solution.Utilize said SIFT characteristic with a high credibility to asking for affine transformation parameter through the heavy weighted least-squares method of iteration, said affined transformation model is as the initial transformation model of smart step of registration.
Fig. 2 is the process flow diagram of the extraction SIFT characteristic of step S2 in the specific embodiment of method of the present invention.In this embodiment, the SIFT Feature Extraction realizes according to following flow process:
The DOG of step S21 ', design of graphics picture (Difference of Gaussian, difference of gaussian) pyramid.
If image be I (x, y), then this difference image of k floor height is D (x, y, k σ)=L (x in the DOG pyramid; Y, (k+1) σ)-L (x, y, k σ); Wherein, and L (x, y, σ)=G (x; Y, σ) (x, y),
* representes convolution algorithm to * I.Fig. 3 is that DOG makes up legend, and as shown in Figure 3, image carries out Gaussian convolution on 5 yardsticks, and 4 images are arranged in the DOG pyramid that obtains.
Step S22 ', on pyramidal every layer of DOG, extract extreme point, so-called extreme point is meant the point that D in local neighborhood (x, y, k σ) value is maximum.
The process of extracting extreme point does, selects any point on the DOG pyramid, if this is not an extreme point in 26 neighborhoods of this layer and upper and lower adjacent two layers, then this point is removed, otherwise with this point as extreme point.Fig. 4 is the synoptic diagram that extracts extreme point; As shown in Figure 4; The point of 26 marks " ● " of the point of the mark " * " of the k tomographic image in the DOG pyramid and k-1 layer, k+1 layer compares, if corresponding D (x, the y of the point of mark " * "; K σ) be maximal value in these 26 neighborhood points, then the point of mark " * " is an extreme point.
Step S23 ', for the extreme point that is extracted, remove the very asymmetric extreme point of local curvature.In this embodiment, calculate the local Hessian matrix H of difference image D, remove the extreme point of condition below satisfying: tr (H)
2/ det (H)＞10, the determinant of det (H) representing matrix H wherein, the mark of tr (H) representing matrix H;
Other locus of sub-pixel, the yardstick of step S24 ', calculating SIFT characteristic, the SIFT characteristic refers to the extreme point that remains.
Suppose SIFT characteristic X=(x, y, σ), x wherein, y, σ are respectively x, y direction coordinate and the scale parameter of extreme point X, this moment, the coordinate of three directions all was a positive number.According to difference of gaussian image D (x; Y; σ) Taylor expansion formula
calculate with SIFT characteristic X be initial point with respect to X be expert at, side-play amount on row and the number of plies; Promptly
wherein D represent the Taylor expansion formula once, D,
and
can be calculated according to method of difference by the pixel of SIFT characteristic X and neighborhood thereof.According to following rule unique point is carried out the sub-pixel interpolation then: if the side-play amount on three directions all less than 0.5 pixel, then this point be exactly unique point
be the sub-pixel extreme point coordinate of being asked; If the side-play amount on a certain direction is more than or equal to 0.5 pixel; As the side-play amount of supposing the x direction is rounded up to a round values a greater than 0.5 pixel with side-play amount, and a and x addition are obtained new SIFT characteristic X2=(x+a; Y; S), then extreme point X2 is operated by above-mentioned steps, the side-play amount on three directions is all less than 0.5.
Step S25 ', confirm said SIFT characteristic principal direction, so-called principal direction be meant with the neighborhood that is characterized as the center with SIFT in the corresponding gradient direction of peak value of gradient orientation histogram.
In this embodiment; With SIFT characteristic X=(x; Y; σ) for the center, be to sample in the neighborhood window of radius with 1.5 σ, calculate Gauss's smoothed image L (x, y; σ) gradient direction of each pixel in above-mentioned neighborhood window
also with the gradient direction of statistics with histogram neighborhood territory pixel, obtains a gradient orientation histogram.Gradient orientation histogram is a kind of statistical graph about gradient direction θ, and its scope is 0～360 degree, wherein per a 10 degree post, 36 posts altogether.The peak value of this gradient orientation histogram has been represented the principal direction of this unique point place neighborhood gradient, promptly as the direction of this SIFT characteristic.Fig. 5 is an exemplary plot of the histogram of gradients of this embodiment of the present invention.Shown that in the figure adopting 7 posts is the example that unique point is confirmed principal direction.In this gradient orientation histogram, when existing another to be equivalent to the peak value of main peak value 80% energy, then this direction is thought the auxilliary direction of this unique point.A unique point may designatedly have a plurality of directions (principal direction, auxilliary direction more than).
Step S26 ', be that the SIFT latent structure describes vector, the so-called vector of describing is meant and is used to portray vector image block statistical nature around this SIFT characteristic, that be made up of gradient orientation histogram.
In this embodiment, at first the coordinate axis of image block around the SIFT characteristic is rotated to be the principal direction of said SIFT characteristic, to guarantee rotational invariance; The fritter that is divided into 4 * 4 pixels then in the window with 16 * 16 pixels around the unique point calculates the gradient orientation histogram of 8 directions of each fritter, and the gradient orientation histogram of each fritter is coupled together the proper vector that forms 128 dimensions; At last, it is normalized to unit length.Fig. 6 is a synoptic diagram for unique point structure description vector of the present invention.
In this embodiment, the heavy weighted least-squares method of iteration is asked for the initial transformation parameter and is realized according to following flow process:
S21 ", try to achieve affined transformation model θ according to the SIFT characteristic with a high credibility of step S23 to utilizing least square method
Pq, adopt ICP (Iterative Closest Point) method through the forward direction coupling with after to obtain new SIFT characteristic to coupling backward right;
The forward direction coupling is a reference picture with the full-colour image, and the ICP method is that each the SIFT characteristic on the multispectral image is sought its SIFT characteristic corresponding on full-colour image.Arbitrary SIFT characteristic c for multispectral image
i=(x
i, y
i, s
i, θ
i), it is at affined transformation model θ
PqThe coordinate of corresponding full-colour image is designated as C down
i, distance C on full-colour image
iNearest SIFT characteristic d
i=(x
i, y
i, s
i, θ
i) be c
iThe SIFT characteristic of coupling.
The back is reference picture to coupling with the multispectral image, and the ICP method is that each the SIFT characteristic on the full-colour image is sought its SIFT characteristic corresponding on multispectral image, and operating process and forward direction coupling are similar.
Step S22 ", reappraise forward direction and back to transformation parameter according to the SIFT characteristic of current coupling to utilizing the heavy weighted least-squares method of iteration; Through the alternately iteration of ICP and parameter estimation, confirm optimal transformation parameter and optimum coupling;
Current transformation model and current SIFT characteristic are to being closely related.For this reason, under the prerequisite of given alternative types, adopt the heavy weighted least-squares method of iteration constantly to adjust the SIFT characteristic of transformation parameter and coupling.Suppose that current transformation model does
(k representes alternative types, and k=1 representes that affined transformation, k=2 represent projective transformation) is for convenient narration, at step S22 " in, after alternative types is given with transformation model
Still be designated as θ
PqFor any multi-spectral image SIFT features
Its full-color image matching SIFT features
iterative reweighted least squares objective function is:
In the following formula,
d
f(p, q; θ
Pq)=| (T (p; θ
Pq)-q)
Tη
q|/s
q,
w
F; iBe the right weight of each SIFT characteristic.Calculate for simplifying, during actual finding the solution above-mentioned objective function is split as two steps: find the solution each SIFT characteristic coupling weight w
D, iWith the renewal transformation parameter, these two step cycle cross-iterations are up to algorithm convergence.Coupling weight w
D, iAccording to current transformation model θ
PqFollowing SIFT characteristic is to (p
i, q
i) matching error calculate,
Wherein,
At current coupling weight w
D, iThe basis on, it is new for optimizing to utilize weighted least-squares method to estimate
Find the solution new transformation parameter.σ
fBe the yardstick normalizing factor of SIFT characteristic, computing formula is following:
Corresponding θ during above-mentioned algorithm convergence
PqAnd θ
QpFor the optimum forward transform model of current matching area with afterwards to transformation model.
Step S3 of the present invention is the registration on original full-colour image and original multispectral image.In this embodiment, smart registration is realized according to following flow process:
S31: initial transformation model conversion.The multiple s that samples that falls according to full-colour image and multispectral image
1And s
2And optimum forward transform model θ
PqCalculate the conversion transformation model of corresponding original resolution full-colour image and multispectral image
Be convenient narration, will in step S3
Be designated as θ
Pq
S32: blocking characteristic extracts, coupling.Full-colour image is divided into the even sub-piece of 1024*1024 pixel, to each sub-piece P wherein, according to initial transformation model θ
PqCan obtain corresponding multispectral image piece M; On image block P and M, extract the SIFT characteristic according to the said method of S21 respectively, it is right to obtain the SIFT characteristic of every antithetical phrase interblock according to the said method of S22 then; The SIFT characteristic of collecting all sub-interblocks is right, obtains the SIFT characteristic pair set S0 of corresponding view picture full-colour image and multispectral image.
S33: exterior point is removed.Utilize initial transformation model θ earlier
PqThe geometrical constraint that provides is removed the exterior point among the SIFT characteristic pair set S0, and concrete steps are following: for certain the SIFT characteristic among the S0 to (a, b), a=(x
1, y
1, s
1, θ
1), b=(x
2, y
2, s
2, θ
2), if (x
1, y
1) and T (b, θ
Pq) between Euclidean distance less than τ, it is right then to keep this SIFT characteristic; Otherwise, from S0, remove the SIFT characteristic to obtaining match point pair set S1.Then, on S set 1, utilize RANSAC (RANdom SAmple Consensus, random sampling consistance) algorithm to remove exterior point, obtain overall SIFT characteristic pair set S.T (b, θ
Pq) expression SIFT characteristic coordinates (x
2, y
2) at transformation model θ
PqUnder coordinate.τ is a threshold value, because in step S34, also will recomputate the right weight of each SIFT characteristic through the mode of iteration, the value of τ is little to the registration accuracy influence.In the instance of the present invention, τ=50.
S34: alternative types is selected and accurate transformation parameter is found the solution.Be final transformation model with affined transformation and projective transformation respectively, utilize SIFT that step S33 obtains carrying out alternative types and select, find the solution transformation parameter through iterating heavy weighted least-squares method.
In this embodiment, alternative types selection and accurate transformation parameter are found the solution according to following flow process and are realized:
Make alternative types k=1; 2; And carry out step S22 operation respectively, obtain corresponding forward transform model and back be respectively forward direction from
and
current optimal coupling to conversion with after to SIFT characteristic pair set be for
and
immoderation under Akaike Information criterion:
In the following formula; The degree of freedom of
1 expression alternative types k; During k=1,1=6; During k=2,1=9; | A| representes element number in the set A.With minimal Ii corresponding
and
the optimal matching area for the current forward and backward transformation model.
The step S4 of this embodiment of the present invention realizes by following flow process: the transformation model of trying to achieve according to step S3; To the multispectral image after the bicubic spline interpolation obtains conversion that carries out of multispectral image, multispectral image after the conversion and full-colour image be registration.
Above-described specific embodiment; The object of the invention, technical scheme and beneficial effect have been carried out further explain, it should be understood that the above is merely specific embodiment of the present invention; Be not limited to the present invention; All within spirit of the present invention and principle, any modification of being made, be equal to replacement, improvement etc., all should be included within protection scope of the present invention.
Claims (7)
1. the method for registering of visible light full-colour image and multispectral image is characterized in that, comprises the steps:
Step S1, full-colour image and multispectral image are carried out multiple dimensioned decomposition, generate the full-colour image and the multispectral image of low resolution;
Step S2, said low-resolution image on extract, coupling SIFT characteristic and remove exterior point, the SIFT characteristic of utilizing coupling is to obtaining the initial transformation model with the heavy weighted least-squares method of iteration;
Step S3, original image on utilize said initial transformation model according to the geometrical constraint that provides; Carry out removing, and on all SIFT characteristic pair sets, select optimum alternative types and utilize the heavy weighted least-squares method of iteration to try to achieve transformation parameter based on the right SIFT feature extraction of image block, coupling and exterior point;
Step S4, multispectral image is carried out conversion, obtain the multispectral image after the conversion according to said transformation model.
2. the method for registering of visible light full-colour image as claimed in claim 1 and multispectral image is characterized in that, said step S1 comprises:
Step S11: if the known image latitude and longitude information, the overlapping region of taking out full-colour image and multispectral image according to latitude and longitude information; If there is not the image latitude and longitude information, utilize the entire image information of two width of cloth images to carry out registration;
Step S12: the multiple dimensioned decomposition number of plies n that confirms multispectral image according to the size of multispectral image
1, confirm the multiple dimensioned decomposition number of plies n of full-colour image according to the full-colour image size and with the spatial resolution difference of multispectral image
2
Step S13: sampling is fallen in full-colour image and multispectral image, and full-colour image falls the sampling multiple and falls the sampling multiple for
for
multispectral image
3. the method for registering of visible light full-colour image as claimed in claim 1 and multispectral image is characterized in that, said step S2 comprises:
S21, on low resolution full-colour image and low resolution multispectral image, extract the SIFT characteristic respectively, obtain two SIFT characteristic sets;
S22, two SIFT characteristic sets are mated based on the Euclidean distance and the arest neighbors ratioing technigue of SIFT feature description vector:
S23, full-colour image under the low resolution and the alternative types between the multispectral image are regarded as affined transformation, utilize the RANSAC algorithm to remove exterior point, it is right to obtain SIFT characteristic with a high credibility;
S24, utilize said SIFT characteristic with a high credibility to asking for affine transformation parameter through the heavy weighted least-squares method of iteration, said affined transformation model is as the initial transformation model of smart step of registration.
4. the method for registering of visible light full-colour image as claimed in claim 3 and multispectral image is characterized in that, the step of the extraction SIFT characteristic of step S2 comprises:
The DOG pyramid of step S21 ', design of graphics picture;
Step S22 ', on pyramidal every layer of DOG, extract extreme point, so-called extreme point is meant the point that D in local neighborhood (x, y, k σ) value is maximum;
Step S23 ', for the extreme point that is extracted, remove the asymmetric extreme point of local curvature;
Other locus of sub-pixel, the yardstick of step S24 ', calculating SIFT characteristic, the SIFT characteristic refers to the extreme point that remains;
Step S25 ', confirm said SIFT characteristic principal direction, so-called principal direction be meant with the neighborhood that is characterized as the center with SIFT in the corresponding gradient direction of peak value of gradient orientation histogram;
Step S26 ', be that the SIFT latent structure describes vector, the so-called vector of describing is meant and is used to portray vector image block statistical nature around this SIFT characteristic, that be made up of gradient orientation histogram.
5. the method for registering of visible light full-colour image as claimed in claim 3 and multispectral image is characterized in that, the step that the heavy weighted least-squares method of the iteration of said step S2 is asked for the initial transformation parameter comprises:
Step S21 ", try to achieve affined transformation model θ according to the SIFT characteristic with a high credibility of step S23 to utilizing least square method
Pq, adopt ICP (Iterative Closest Point) method through the forward direction coupling with after to obtain new SIFT characteristic to coupling backward right;
Step S22 ", reappraise forward direction and back to transformation parameter according to the SIFT characteristic of current coupling to utilizing the heavy weighted least-squares method of iteration; Through the alternately iteration of ICP and parameter estimation, confirm optimal transformation parameter and optimum coupling;
6. the method for registering of visible light full-colour image as claimed in claim 1 and multispectral image is characterized in that, said step S3 comprises:
Step S31, according to full-colour image and multispectral image sampling multiple s falls
1And s
2And optimum forward transform model θ
PqCalculate the conversion transformation model of corresponding original resolution full-colour image and multispectral image
Step S32, full-colour image is divided into even sub-piece,, obtains corresponding multispectral image piece M according to the initial transformation model to each sub-piece P wherein; On image block P and M, extract the SIFT characteristic respectively, the SIFT characteristic that obtains every antithetical phrase interblock then is right, and the SIFT characteristic of collecting all sub-interblocks is right, obtains the SIFT characteristic pair set of corresponding view picture full-colour image and multispectral image;
Step S33, utilize initial transformation model θ
PqThe geometrical constraint that provides is removed the exterior point in the SIFT characteristic pair set, then, in this set, utilizes the RANSAC algorithm to remove exterior point, obtains overall SIFT characteristic pair set;
Step S34, be final transformation model with affined transformation and projective transformation respectively, the SIFT characteristic of utilizing step S33 to obtain is right, carries out alternative types and selects, finds the solution transformation parameter through iterating heavy weighted least-squares method;
7. the method for registering of visible light full-colour image as claimed in claim 1 and multispectral image is characterized in that, said step S4 comprises:
The multispectral image after the bicubic spline interpolation obtains conversion that carries out to multispectral image.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201210251487.1A CN102800098B (en) | 2012-07-19 | 2012-07-19 | Multi-characteristic multi-level visible light full-color and multi-spectrum high-precision registering method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201210251487.1A CN102800098B (en) | 2012-07-19 | 2012-07-19 | Multi-characteristic multi-level visible light full-color and multi-spectrum high-precision registering method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN102800098A true CN102800098A (en) | 2012-11-28 |
| CN102800098B CN102800098B (en) | 2015-03-11 |
Family
ID=47199196
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201210251487.1A Expired - Fee Related CN102800098B (en) | 2012-07-19 | 2012-07-19 | Multi-characteristic multi-level visible light full-color and multi-spectrum high-precision registering method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN102800098B (en) |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103065311A (en) * | 2012-12-26 | 2013-04-24 | 中国土地勘测规划院 | Satellite image automatic registration method based on standard image database |
| CN103325127A (en) * | 2013-05-15 | 2013-09-25 | 深圳大学 | Method for extracting and describing multi-spectral image SIFT features and system of method |
| CN104318583A (en) * | 2014-11-14 | 2015-01-28 | 武汉大学 | Visible light broadband spectrum image registration method |
| CN104966283A (en) * | 2015-05-22 | 2015-10-07 | 北京邮电大学 | Imaging layered registering method |
| CN105503625A (en) * | 2016-01-15 | 2016-04-20 | 中国海洋石油总公司 | Method for synthesizing diethanolamine acrylate compounds |
| CN105523949A (en) * | 2016-01-15 | 2016-04-27 | 中国海洋石油总公司 | Method for preparing triethanolamine acrylate compound |
| CN105844630A (en) * | 2016-03-21 | 2016-08-10 | 西安电子科技大学 | Binocular visual image super-resolution fusion de-noising method |
| CN106611410A (en) * | 2016-11-29 | 2017-05-03 | 北京空间机电研究所 | Pansharpen fusion optimization method based on pyramid model |
| CN108932708A (en) * | 2018-06-13 | 2018-12-04 | 中国科学院光电研究院 | Satellite-borne multispectral remote sensing image fusion method based on super-resolution enhancing |
| CN110660089A (en) * | 2019-09-25 | 2020-01-07 | 云南电网有限责任公司电力科学研究院 | Satellite image registration method and device |
| CN110660038A (en) * | 2019-09-09 | 2020-01-07 | 山东工商学院 | Multispectral image and panchromatic image fusion method based on generation countermeasure network |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2007072451A2 (en) * | 2005-12-22 | 2007-06-28 | Philips Intellectual Property & Standards Gmbh | Adaptive point-based elastic image registration |
| CN101989352A (en) * | 2009-08-06 | 2011-03-23 | 宋智礼 | Image registration method based on improved scale invariant feature transform (SIFT) algorithm and Lissajous figure track |
-
2012
- 2012-07-19 CN CN201210251487.1A patent/CN102800098B/en not_active Expired - Fee Related
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2007072451A2 (en) * | 2005-12-22 | 2007-06-28 | Philips Intellectual Property & Standards Gmbh | Adaptive point-based elastic image registration |
| CN101989352A (en) * | 2009-08-06 | 2011-03-23 | 宋智礼 | Image registration method based on improved scale invariant feature transform (SIFT) algorithm and Lissajous figure track |
Non-Patent Citations (3)
| Title |
|---|
| CHUNLEI HUO ET AL.: "Multilevel SIFT Matching for Large-Size VHR Image Registration", 《IEEE GEOSCIENCE AND REMOTE SENSING LETTERS》 * |
| DAVID G. LOWE: "Distinctive Image Features from Scale-Invariant Keypoints", 《INTERNATIONAL JOURNAL OF COMPUTER》 * |
| 丁琳等: "基于不变特征的CCD 影像与红外影像的自动配准", 《红外与激光工程》 * |
Cited By (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103065311A (en) * | 2012-12-26 | 2013-04-24 | 中国土地勘测规划院 | Satellite image automatic registration method based on standard image database |
| CN103325127B (en) * | 2013-05-15 | 2016-01-20 | 深圳大学 | A kind of multispectral image SIFT feature is extracted and describing method and system thereof |
| CN103325127A (en) * | 2013-05-15 | 2013-09-25 | 深圳大学 | Method for extracting and describing multi-spectral image SIFT features and system of method |
| CN104318583B (en) * | 2014-11-14 | 2017-02-22 | 武汉大学 | Visible light broadband spectrum image registration method |
| CN104318583A (en) * | 2014-11-14 | 2015-01-28 | 武汉大学 | Visible light broadband spectrum image registration method |
| CN104966283A (en) * | 2015-05-22 | 2015-10-07 | 北京邮电大学 | Imaging layered registering method |
| CN105503625A (en) * | 2016-01-15 | 2016-04-20 | 中国海洋石油总公司 | Method for synthesizing diethanolamine acrylate compounds |
| CN105523949A (en) * | 2016-01-15 | 2016-04-27 | 中国海洋石油总公司 | Method for preparing triethanolamine acrylate compound |
| CN105844630A (en) * | 2016-03-21 | 2016-08-10 | 西安电子科技大学 | Binocular visual image super-resolution fusion de-noising method |
| CN105844630B (en) * | 2016-03-21 | 2018-11-16 | 西安电子科技大学 | A kind of image super-resolution fusion denoising method of binocular vision |
| CN106611410A (en) * | 2016-11-29 | 2017-05-03 | 北京空间机电研究所 | Pansharpen fusion optimization method based on pyramid model |
| CN106611410B (en) * | 2016-11-29 | 2019-10-22 | 北京空间机电研究所 | Pansharpen based on pyramid model merges optimization method |
| CN108932708A (en) * | 2018-06-13 | 2018-12-04 | 中国科学院光电研究院 | Satellite-borne multispectral remote sensing image fusion method based on super-resolution enhancing |
| CN110660038A (en) * | 2019-09-09 | 2020-01-07 | 山东工商学院 | Multispectral image and panchromatic image fusion method based on generation countermeasure network |
| CN110660038B (en) * | 2019-09-09 | 2023-06-09 | 山东工商学院 | Multispectral image and full-color image fusion method based on generation countermeasure network |
| CN110660089A (en) * | 2019-09-25 | 2020-01-07 | 云南电网有限责任公司电力科学研究院 | Satellite image registration method and device |
Also Published As
| Publication number | Publication date |
|---|---|
| CN102800098B (en) | 2015-03-11 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN102800098A (en) | Multi-characteristic multi-level visible light full-color and multi-spectrum high-precision registering method | |
| Wu et al. | Pointpwc-net: A coarse-to-fine network for supervised and self-supervised scene flow estimation on 3d point clouds | |
| CN102800097A (en) | Multi-feature multi-level visible light and infrared image high-precision registering method | |
| CN104574347B (en) | Satellite in orbit image geometry positioning accuracy evaluation method based on multi- source Remote Sensing Data data | |
| CN103034982B (en) | Image super-resolution rebuilding method based on variable focal length video sequence | |
| CN100435167C (en) | High resolution SAR image registration processing method and system | |
| CN102819839B (en) | High-precision registration method for multi-characteristic and multilevel infrared and hyperspectral images | |
| CN103727930B (en) | A kind of laser range finder based on edge matching and camera relative pose scaling method | |
| CN104200461B (en) | The remote sensing image registration method of block and sift features is selected based on mutual information image | |
| CN110689622A (en) | Synchronous positioning and composition algorithm based on point cloud segmentation matching closed-loop correction | |
| CN105389774A (en) | Method and device for aligning images | |
| CN102800099A (en) | Multi-feature multi-level visible light and high-spectrum image high-precision registering method | |
| CN101882308A (en) | Method for improving accuracy and stability of image mosaic | |
| CN105335965B (en) | Multi-scale self-adaptive decision fusion segmentation method for high-resolution remote sensing image | |
| CN102629374A (en) | Image super resolution (SR) reconstruction method based on subspace projection and neighborhood embedding | |
| CN111540005A (en) | Loop detection method based on two-dimensional grid map | |
| CN112052783A (en) | High-resolution image weak supervision building extraction method combining pixel semantic association and boundary attention | |
| CN102446356A (en) | Parallel and adaptive matching method for acquiring remote sensing images with homogeneously-distributed matched points | |
| CN104063711A (en) | Corridor vanishing point rapid detection algorithm based on K-means method | |
| CN102122359A (en) | Image registration method and device | |
| CN103218811A (en) | Statistical distribution-based satellite multi-spectral image waveband registration method | |
| Li et al. | 3d building reconstruction from monocular remote sensing images | |
| CN103679740B (en) | ROI (Region of Interest) extraction method of ground target of unmanned aerial vehicle | |
| CN113409332B (en) | Building plane segmentation method based on three-dimensional point cloud | |
| CN109242832B (en) | Multi-temporal multispectral remote sensing image change detection method and system |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20150311 Termination date: 20210719 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |