CN105427298B - Remote sensing image registration method based on anisotropic gradient metric space - Google Patents
Remote sensing image registration method based on anisotropic gradient metric space Download PDFInfo
- Publication number
- CN105427298B CN105427298B CN201510770880.5A CN201510770880A CN105427298B CN 105427298 B CN105427298 B CN 105427298B CN 201510770880 A CN201510770880 A CN 201510770880A CN 105427298 B CN105427298 B CN 105427298B
- Authority
- CN
- China
- Prior art keywords
- remote sensing
- sensing images
- registration
- subject
- metric space
- 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
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10032—Satellite or aerial image; Remote sensing
Abstract
The invention discloses a kind of remote sensing image registration method based on anisotropic gradient metric space, mainly solve the problems, such as that it is relatively low to change greatly correct matching rate during situation for luminance non-linearity between remote sensing images.Implementation step is：(1) remote sensing images pair are inputted；(2) metric space of structural anisotropy's diffusion；(3) gradient amplitude image is calculated；(4) characteristic point is detected；(5) characteristic point principal direction is generated；(6) feature point description is generated；(7) Feature Points Matching；(8) characteristic point pair of deletion error matching；(9) registering reference picture and image subject to registration.The present invention carries out feature point detection, the generation of characteristic point principal direction and feature point description son generation on the gradient amplitude image of anisotropy metric space, brightness between image, which can be successfully managed, the situation of larger heterogeneous linear change, available for complicated multi-source and multi-spectral remote sensing image registration.
Description
Technical field
The invention belongs to technical field of image processing, further relates to carry out registration process technical field to remote sensing images
In a kind of remote sensing image registration method based on anisotropic gradient metric space.Present invention can apply to multispectral, more
The registration for the remote sensing images that source obtains at different moments.
Background technology
Image registration refers in different periods, and Same Scene is shot using identical or different sensors with different view
Have the process that overlapping region image carries out Geometry rectification.Image registration techniques are the image processing techniques quickly grown in recent years
One of.Image registration techniques are widely used for every field, such as aeronautical and space technology, image mosaic, GIS-Geographic Information System, figure
As fusion, three-dimensional reconstruction, target identification and change detection etc..Now with the development of remote sensing technology, due to sensor not
Be on the increase with remote sensing images caused by physical characteristic, thus comprehensively utilize various images carry out data extractions and analysis into
For an important means of remote sensing fields.Simultaneously because physical characteristic and the difference of imaging mode between various sensors, in number
According to strict registration must be carried out between the image of difference geometrical property when application and data fusion and different resolution.
Remote sensing image registration is broadly divided into two classes at present：The registration side of method for registering and feature based based on area grayscale
Method.Wherein, the conventional method for registering images based on area grayscale has：Cross-correlation (CC), not bending moment, the phase based on FFT are related
Method and mutual information (MI) etc..But based on the method for registering images of gray scale, there is following shortcoming：1. to variation of image grayscale ratio
More sensitive, especially nonlinear illumination variation, this greatly reduces the performance of algorithm；2. computation complexity is high；3. to target
Rotation, deformation and to block comparison sensitive.The shortcomings that in order to overcome it to exist, there has been proposed the image registration side of feature based
Method.Method based on characteristics of image extracts the features such as edge, angle point, profile and regional center first from image, then uses
The related optimal alignment for determining image between feature.These notable features can greatly reduce the information content of image so that meter
Calculation amount reduces, and speed faster, and has robustness to the grey scale change of image.Point feature is the most commonly used and imitated at present
A kind of higher method of rate.Angle point, flex point, crosspoint in two dimensional image etc. are the obvious characteristics of image.These points have flat
Consistency is moved, rotated and scaled, is hardly influenceed by illumination condition.It only need to use image in about 0.05% pixel just
The data message of entire image can be represented, its information content is high, calculating speed is fast, makes to be treated as possibility in real time.Therefore, base
Also it is widely used in remote sensing image registration field in the method for registering images of feature.
Classical SIFT algorithms have keeps constant to visible images rotation, scaling, the illumination of part and view transformation
The advantages of property.When SIFT methods detection feature counts out increase, generation characteristic point principal direction and generation feature point description sub-step
It is rapid to spend the time to increase sharply, limit the application of SIFT algorithms in practice.The advantages of due to SIFT algorithms, domestic and foreign scholars
Propose many boosting algorithms based on SIFT.But there is many when matching remote sensing images using the method based on SIFT
Incorrect match point, correct matching rate (CMT) decline rapidly.Reason is shooting time, spectrum and obtains sensor not
It is same to cause remote sensing images dramatically different to the pixel intensity of the same area, while the brightness mapping between pixel pair is probably line
It is property, non-linear, or even brightness Reversion occurs.
Paper " the Uniform Robust Scale-Invariant Feature that Sedaghat delivers at it
Matching for Optical Remote Sensing Images”(《IEEE Transactions on Geoscience
and Remote Sensing》, 2011, pages 4516-4527) in propose a kind of UR-SIFT algorithms.The algorithm is preferable
Quantity, quality and the distribution of the characteristic point of metric space extraction are controlled, improves the part remote sensing images with partial transformation
Registration accuracy, but the weak point that the algorithm still has is, it is impossible to which accurate registering brightness of image has larger non-linear
The remote sensing images pair of change.
Paper " the A Novel Coarse-to-Fine Scheme for Automatic Image that Gong delivers at it
Registration Based on SIFT and Mutual Information”(《IEEE Transactions on
Geoscience and Remote Sensing》, 2014, pages 4328-4338) in disclose and a kind of being matched somebody with somebody by thick to smart
Quasi- method.This method first obtains the initial transformation relation between image pair using SIFT algorithms, then in conjunction with initial transformation relation
More accurate image registration is obtained with mutual information method.The weak point that this method still has is, because this method is base
In the method for mutual information, so computation complexity is high, and registration accuracy works as image according to the initial solution for being disinclined to the acquisition of sift algorithms
In the case that brightness has larger nonlinear change, sift algorithms can not obtain good initial solution, rely on sift algorithms and obtain
The mutual information method of the initial solution obtained can not realize accurate registration.
The patent that Xian Electronics Science and Technology University applies at it " selects block and the remote sensing image registration of sift features based on mutual information
Method " (number of patent application：CN201410379927, publication number：CN104200461A proposed in) a kind of based on sift and mutual
The remote sensing image registration method of information.The implementation process of this method is：At random from reference to remote sensing images and remote sensing images subject to registration
Middle selection image pair, calculate the mutual information of every a pair of images；Descending arrangement is carried out to mutual information；N mutual information is larger before selection
Subgraph pair；The larger image of preceding n mutual information to selection is slightly matched to extracting sift features；Deletion error
Match point, carefully matched；Calculate registration parameter and association relationship；The maximum registration parameter of mutual information is chosen as final to match somebody with somebody
Quasi- parameter.Although this method can accelerate the speed of image registration, the weak point that this method still has is, in spy
The descriptor generation phase of the generation of sign point, the generation of characteristic point principal direction and characteristic point still uses sift methods, in remote sensing images pair
Luminance non-linearity change greatly in the case of correct matching rate decline rapidly, while this method using traditional arest neighbors and time
Nearest neighbor distance than matching criterior, count out also fast by the feature correctly matched in the case of it more repeated characteristic in remote sensing images be present
Speed declines.
The content of the invention
The purpose of the present invention is to overcome the shortcomings of above-mentioned prior art, proposes that one kind is based on anisotropic gradient metric space
Remote sensing image registration method, to improve the degree of accuracy of characteristic matching, realizing has the distant of larger nonlinear change to brightness of image
Feel the registration of image pair.
The present invention realizes that the thinking of above-mentioned purpose is：With reference to remote sensing images and treated according to Nonlinear Diffusion principle structure first
The anisotropy metric space of registering remote sensing images, then to the anisotropy yardstick with reference to remote sensing images and remote sensing images subject to registration
On the gradient amplitude image in space using Harris operators carry out Corner Detection, then generate point feature to the non-of brightness of image
Linear change has description of higher robustness, finally carries out Feature Points Matching using improved Feature Points Matching criterion, obtains
Last remote sensing images are to registration result figure.
The step of the present invention includes as follows：
(1) input refers to remote sensing images and remote sensing images subject to registration；
(2) metric space of structural anisotropy's diffusion：
(2a) calculates the scale-value of each layer of anisotropy metric space with reference to remote sensing images and remote sensing images subject to registration；
Scale-value is transformed into time measure value by (2b)；
The reference remote sensing images and remote sensing images subject to registration of (2c) to input, use standard deviation as σ0Gaussian filtering, obtain
To with reference to remote sensing images and the tomographic image of remote sensing images anisotropy metric space the 0th subject to registration；
The sequence number i of anisotropy metric space layer is initialized as zero by (2d)；
(2e) according to the following formula, calculates the with reference to remote sensing images and remote sensing images anisotropy metric space subject to registration respectively
The diffusion coefficient matrix of i tomographic images：
Wherein, ciRepresent the i-th tomographic image with reference to remote sensing images or remote sensing images anisotropy metric space subject to registration
Diffusion coefficient matrix,Represent i-th layer of figure with reference to remote sensing images or remote sensing images anisotropy metric space subject to registration
As IiThe image after gaussian filtering for being 1 using standard deviation,Represent the image after gaussian filteringGradient amplitude, | |
Modulo operation is represented, K represents contrast factor, and K value is gradient amplitudeThe hundredths of statistic histogram 70% ladder
Spend range value；
(2f) according to the following formula, schemes to i-th layer with reference to remote sensing images and remote sensing images anisotropy metric space subject to registration
As doing capable diffusion：
Wherein,Represent along i-th layer of figure with reference to remote sensing images or remote sensing images anisotropy metric space subject to registration
As IiRow diffusion after image, I1Represent the columns of line number and columns with referring to remote sensing images or remote sensing images subject to registration
Identical unit square formation, tiAnd ti+1Represent to refer to remote sensing images or remote sensing images anisotropy metric space subject to registration respectively
I-th layer and i+1 layer time measure value, A1(Ii) presentation code with reference to remote sensing images or remote sensing images subject to registration respectively to
I-th tomographic image diffusion coefficient c of different in nature metric spaceiMatrix, IiRepresent each with reference to remote sensing images or remote sensing images subject to registration
I-th tomographic image of anisotropy metric space, ()-1Represent inverse matrix operation；
(2g) according to the following formula, schemes to i-th layer with reference to remote sensing images and remote sensing images anisotropy metric space subject to registration
As doing row diffusion：
Wherein,Represent along i-th layer of figure with reference to remote sensing images or remote sensing images anisotropy metric space subject to registration
As IiRow diffusion after image, I2Represent the line number of line number and columns with reference remote sensing images or remote sensing images subject to registration
Identical unit square formation；
(2h) according to the following formula, calculates the i+1 with reference to remote sensing images and remote sensing images anisotropy metric space subject to registration
Tomographic image：
Wherein, Ii+1Represent the i+1 layer figure with reference to remote sensing images or remote sensing images anisotropy metric space subject to registration
Picture；
(2i) judges whether i >=N-2 sets up, if so, obtain with reference to remote sensing images and remote sensing images subject to registration it is each to
Different in nature metric space, otherwise, i=i+1 is made, perform step (2e)；Wherein, N represents to refer to remote sensing images and remote sensing figure subject to registration
As the sum of anisotropy metric space layer；
(3) gradient amplitude image is calculated：
Using Sobel Operator Sobel, calculating refers to remote sensing images and remote sensing images anisotropy metric space subject to registration
Gradient amplitude image；
(4) difference of gradient amplitude image is calculated：
(4a) according to the following formula, using Sobel Operator Sobel, calculate with reference to remote sensing images and remote sensing images subject to registration respectively to
The difference of the x-axis positive direction of the gradient amplitude image of different in nature metric space：
Wherein,Represent the gradient amplitude with reference to remote sensing images or remote sensing images anisotropy metric space subject to registration
Image GInThe horizontally difference of right x-axis positive direction, GInRepresent with reference to remote sensing images or remote sensing images subject to registration respectively to different
The n-th layer image of the gradient amplitude image of property metric space,Represent associative operation, n=0,1 ..., N-1, N represent with reference to distant
Feel the sum of image and remote sensing images anisotropy metric space layer subject to registration；
(4b) according to the following formula, using Sobel Operator Sobel, calculate with reference to remote sensing images and remote sensing images subject to registration respectively to
The difference of the y-axis positive direction of the gradient amplitude image of different in nature metric space：
Wherein,Represent the gradient amplitude with reference to remote sensing images or remote sensing images anisotropy metric space subject to registration
Image GInAlong the difference of y-axis positive direction straight down；
(5) according to the following formula, the gradient amplitude of gradient amplitude image is calculated：
Wherein, GGInRepresent the gradient amplitude with reference to remote sensing images or remote sensing images anisotropy metric space subject to registration
The n-th layer image GI of imagenGradient amplitude,Radical sign operation is opened in expression；
(6) according to the following formula, the gradient angle of gradient amplitude image is calculated：
Wherein, AGGInRepresent the gradient width with reference to remote sensing images or remote sensing images anisotropy metric space subject to registration
Spend the n-th layer image GI of imagenGradient angle, arctan () represents the operation of four-quadrant arc tangent；
(7) characteristic point is detected：
It is empty in the anisotropy yardstick with reference to remote sensing images and remote sensing images subject to registration using Harris corner detection operators
Between gradient amplitude image on detect characteristic point, obtain set of characteristic points：
Wherein, CIRRepresent the feature detected on the gradient amplitude image with reference to the anisotropy metric space of remote sensing images
Point set, CISRepresent the feature point set detected on the gradient amplitude image of the anisotropy metric space of remote sensing images subject to registration
Close, R represents the sum of the characteristic point detected on the gradient amplitude image with reference to the anisotropy metric space of remote sensing images, S
Represent the sum of characteristic point detected on the gradient amplitude image of the anisotropy metric space of remote sensing images subject to registration；
(8) characteristic point principal direction is generated：
(8a) 36 deciles in [0,2 π] by horizontal direction angle；
(8b) is 6 σ by the radius of set of characteristic pointsnCircle shaped neighborhood region region in pixel gradient direction AGGIn(X), it is determined that
Decile angular range in border circular areas where pixel, wherein, σnRepresent each with reference to remote sensing images or remote sensing images subject to registration
The scale-value of the n-th layer of anisotropy metric space, AGGIn(X) represent that position coordinates is X picture in circle shaped neighborhood region around characteristic point
The gradient angle of element；
The gradient magnitude GGI for all pixels that (8c) adds up in the range of each decile angle respectivelyn(X), formed
Gradient orientation histogram, wherein, GGIn(X) represent that position coordinates is the gradient width of X pixel in circle shaped neighborhood region around characteristic point
Degree；
(8d) will be more than gradient direction corresponding to the numerical value of 0.8 times of maximum in gradient orientation histogram, as characteristic point
Principal direction；
(9) feature point description is generated：
The circle shaped neighborhood region rotate counterclockwise θ of (9a) by radius for ρ characteristic pointmDegree, circle shaped neighborhood region is along radial direction by ρ points
For 3 sections, the inner circle radius of neighbourhood is 0.25 ρ, and the middle circle radius of neighbourhood is 0.73 ρ, and the cylindrical radius of neighbourhood is ρ, circle shaped neighborhood region
[0,2 π] is divided into 8 sections along angle direction, inner circle as an entirety, around characteristic point circle shaped neighborhood region be divided in order to
The subregion of the area equation of 17 diverse locations, wherein, ρ value is 12 σn, θmRepresent the principal direction of this feature point；
The cartesian coordinate of pixel in characteristic point circle shaped neighborhood region is converted to log-polar, log-polar angle by (9b)
Direction level to the right, is divided into 8 sections in the range of [0,2 π], log-polar logarithm length direction straight down,In the range of it is non-be divided into 3 sections, wherein, ρ represent characteristic point around circle shaped neighborhood region radius；
All pixels in (9c) log-polar grid in every sub-regions are according to its gradient amplitude GGInAnd gradient (X)
Direction AGGIn(X) gradient orientation histogram is calculated, each sub-regions form the gradient direction vector of one 8 dimension, spelled successively
The description for the characteristic point that the gradient direction vector for connecing 17 subregions is formed one 136 dimension is sub, wherein, gradient orientation histogram
Angle 8 sections are divided into the range of [0,2 π]；
(9d) is using step (9a), step (9b), the same procedure of step (9c), generation set of characteristic points CIRDescription son
Set DIRWith set of characteristic points CISDescription subclass DIS,Its
In, DIRRepresent description of characteristic point detected on the gradient amplitude image with reference to the anisotropy metric space of remote sensing images
Set, DISThe characteristic point for representing to detect on the gradient amplitude image of the anisotropy metric space of remote sensing images subject to registration is retouched
State subclass, L1The characteristic point that expression detects on the gradient amplitude image with reference to the anisotropy metric space of remote sensing images
Description son sum, L2Represent the feature detected on the gradient amplitude image of the anisotropy metric space of remote sensing images subject to registration
The description son sum of point；
(10) Feature Points Matching：
The sub- sequence number k of description of Characteristics of The Remote Sensing Images point subject to registration is initialized as 1 by (10a)；
(10b) according to the following formula, calculates description of Characteristics of The Remote Sensing Images point subject to registrationWith with reference to remote sensing images each
The vector of the Euclidean distance composition of description of characteristic point：
Wherein, dkRepresent description of Characteristics of The Remote Sensing Images point subject to registrationWith each characteristic point with reference to remote sensing images
Description son Euclidean distance composition vector,dk,jRepresent description of registering Characteristics of The Remote Sensing Images pointWith description with reference to Characteristics of The Remote Sensing Images pointEuclidean distance,Represent the kth of Characteristics of The Remote Sensing Images point subject to registration
Individual description,Represent j-th of description with reference to Characteristics of The Remote Sensing Images point, j=1,2 ..., L1, L1Expression refers to remote sensing figure
As feature point description son number, | | | | expression take norm to operate；
(10c) is to vectorial dkIn L1Individual element sorts to obtain vectorial s from small to largek；
(10d) ifThen descriptionCorresponding characteristic point and with description sonEuclidean distance be sk,0's
The Feature Points Matching with reference to corresponding to description of remote sensing images；IfAndThen descriptionCorresponding spy
Levy point and with describing sonEuclidean distance be dkPreceding E element reference remote sensing images description corresponding to characteristic point
Match somebody with somebody；IfThen descriptionAny Feature Points Matching of the corresponding characteristic point discord with reference to remote sensing images；
Wherein, sk,0Represent vectorial skFirst element, sk,1Represent vectorial skSecond element, TLAnd THRepresent most
Neighbour and time nearest neighbor distance ratio threshold value, E span are the integer in [2,3]；
(10e) judges k >=L2Whether set up, if so, obtain remote sensing images subject to registration and the feature with reference to remote sensing images
Point matching relation CIR,IS,Perform step (11)；Otherwise, k=k+1 is made, performs step
(10b)；
Wherein, CIR,ISRepresent the spy detected on the gradient amplitude image with reference to the anisotropy metric space of remote sensing images
The pairing of sign point and the characteristic point detected on the gradient amplitude image of the anisotropy metric space of remote sensing images subject to registration
Set of relationship, L3Represent the point of Feature Points Matching stage match to sum；
(11) characteristic point pair of deletion error matching：
Using the consistent RANSAC algorithms of random sampling, the matching double points of deletion error, remote sensing images subject to registration are obtained to ginseng
Examine the geometric transform relation of remote sensing images；
(12) registering reference picture and image subject to registration：
According to geometric transform relation, remote sensing images and remote sensing images subject to registration are referred to using bilinear interpolation method registration.
The present invention has the following advantages that compared with prior art：
First, because the present invention carries out feature point detection on the gradient amplitude image of anisotropy metric space, use
The gradient amplitude of the gradient amplitude image of anisotropy metric space and the principal direction and characteristic point of gradient angle calculation characteristic point
Descriptor, the problem of luminance non-linearity that overcoming prior art can not tackle between remote sensing images pair changes greatly so that
The present invention improves the correct matching rate of remote sensing images pair.
Second, because in the principal direction and generation feature point description sub-stage of generation characteristic point and Gauss is not used in the present invention
Weight window, it is high in generation characteristic point principal direction and generation feature point description sub-stage computation complexity to overcome prior art
Problem so that the present invention improves the operational efficiency of algorithm.
3rd, because the present invention in the Feature Points Matching stage is provided with two threshold values to arest neighbors and time nearest neighbor distance ratio,
The feature correctly matched when prior art carries out Feature Points Matching using a threshold value is overcome to count out the problem of few so that this
Invention improves the feature correctly matched and counted out and registration accuracy.
Brief description of the drawings
Fig. 1 is the flow chart of the present invention；
Fig. 2 is circle shaped neighborhood region region division and log-polar grid schematic diagram around the characteristic point of prior art；
Fig. 3 is the consistent RANSAC algorithm flow charts of stochastical sampling of step 11 of the present invention；
Fig. 4 is the first group of multi-spectral remote sensing image pair and registration result figure of emulation experiment of the present invention；
Fig. 5 is the second group of multi-spectral remote sensing image pair and registration result figure of emulation experiment of the present invention；
Fig. 6 is the multi-source Remote Sensing Images pair and registration result figure of emulation experiment of the present invention.
Embodiment
The present invention is described in further detail below in conjunction with the accompanying drawings.
Referring to the drawings 1, step of the invention is as follows.
Step 1, remote sensing images pair are inputted.
Input refers to remote sensing images and remote sensing images subject to registration.
Step 2, the metric space of structural anisotropy's diffusion.
(2a) calculates the scale-value of each layer of anisotropy metric space with reference to remote sensing images and remote sensing images subject to registration：
σe(m, u)=σ02m+u/L
Wherein, σeRepresent the chi of the e layers with reference to remote sensing images or remote sensing images anisotropy metric space subject to registration
Angle value, σ0Expression refers to remote sensing images or the station meter angle value with remote sensing images subject to registration, by different σeThe anisotropy of composition
The common G groups of metric space, every group of L layer, m represent anisotropy metric space group index, m=0,1 ..., G-1, u represent every group
The index of internal layer, u=0,1 ..., L-1, e=0,1,2 ..., N-1.
Scale-value is transformed into time measure value by (2b)：
Wherein, teRepresent the e layers with reference to remote sensing images or remote sensing images anisotropy metric space subject to registration when
Between metric, σeRepresent the yardstick of the e layers with reference to remote sensing images or remote sensing images anisotropy metric space subject to registration, e
=0,1,2 ..., N-1, N represent sum with reference to remote sensing images and remote sensing images anisotropy metric space layer subject to registration, m tables
Show the group index of anisotropy metric space, m=0,1 ..., G-1, G represent each with reference to remote sensing images and remote sensing images subject to registration
The sum of anisotropy metric space group, u represent the index of every group of internal layer, u=0,1 ..., L-1, L represent with reference to remote sensing images and
The sum of every group of layer of remote sensing images anisotropy metric space subject to registration.
The reference remote sensing images and remote sensing images subject to registration of (2c) to input, use standard deviation as σ0Gaussian filtering, obtain
To with reference to remote sensing images and the tomographic image of remote sensing images anisotropy metric space the 0th subject to registration.
The sequence number i of anisotropy metric space layer is initialized as zero by (2d).
(2e) according to the following formula, calculates the with reference to remote sensing images and remote sensing images anisotropy metric space subject to registration respectively
The diffusion coefficient matrix of i tomographic images：
Wherein, ciRepresent the i-th tomographic image with reference to remote sensing images or remote sensing images anisotropy metric space subject to registration
Diffusion coefficient matrix,Represent i-th layer of figure with reference to remote sensing images or remote sensing images anisotropy metric space subject to registration
As IiThe image after gaussian filtering for being 1 using standard deviation,Represent the image after gaussian filteringGradient amplitude, | |
Modulo operation is represented, K represents contrast factor, and K value is gradient amplitudeThe hundredths of statistic histogram 70% ladder
Spend range value.
(2f) is to the i-th tomographic image I with reference to remote sensing images and remote sensing images anisotropy metric space subject to registrationiRow do
One-dimensional diffusion：
The first step：I-th tomographic image I of calculation code anisotropy metric spaceiDiffusion coefficient matrix ciThe square of h rows
Battle array A1,h(Ii)。
Wherein, A1,h(Ii) presentation code is with reference to remote sensing images or remote sensing images anisotropy metric space subject to registration
I-th tomographic image IiDiffusion coefficient matrix ciThe matrix of h rows, ciRepresent each with reference to remote sensing images or remote sensing images subject to registration
I-th tomographic image I of anisotropy metric spaceiDiffusion coefficient matrix,Represent diffusion coefficient matrix ciH rows, J row
Element, J=0,1 ..., Q-1,Representing matrix ciH rows, the set of the row coordinate of the neighborhood of left and right 2 of the element of w row, w
=0,1 ..., Q-1, Q represent columns with reference to remote sensing images or remote sensing images subject to registration.
Second step：According to the following formula, using Thomas algorithm solution anisotropy row diffusion equations.
Wherein,Represent along i-th layer with reference to remote sensing images or the anisotropy metric space of remote sensing images subject to registration
Image IiThe diffusion of h rows after result, I is the columns phase of size and input with reference to remote sensing images or remote sensing images subject to registration
Same unit matrix, tiAnd ti+1It is i-th with reference to remote sensing images or remote sensing images anisotropy metric space subject to registration respectively
The time measure value of layer and i+1 layer, A1,h(Ii) presentation code refers to remote sensing images or remote sensing images anisotropy subject to registration
I-th tomographic image I of metric spaceiDiffusion coefficient matrix ciThe matrix of h rows,Expression refers to remote sensing images or subject to registration
I-th tomographic image I of the anisotropy metric space of remote sensing imagesiH rows, ()-1Represent inverse matrix operation.
3rd step：For h=0,1 ..., P-1, the first step and second step in this step are repeated, is obtainedIts
In, Represent along empty with reference to remote sensing images or remote sensing images anisotropy yardstick subject to registration
Between the i-th tomographic image IiRow do the result after one-dimensional diffusion, P represents the row with reference to remote sensing images or remote sensing images subject to registration
Number.
(2g) is to the i-th tomographic image I with reference to remote sensing images and remote sensing images anisotropy metric space subject to registrationiRow do
One-dimensional diffusion：
The first step：I-th tomographic image I of calculation code anisotropy metric spaceiDiffusion coefficient matrix ciThe square of v row
Battle array A2,v(Ii)。
Wherein, A2,v(Ii) presentation code is with reference to remote sensing images or remote sensing images anisotropy metric space subject to registration
I-th tomographic image IiDiffusion coefficient matrix ciThe matrix of v row, ciRepresent each with reference to remote sensing images or remote sensing images subject to registration
I-th tomographic image I of anisotropy metric spaceiDiffusion coefficient matrix,Representing matrix ciM rows, v row element, M=
0,1 ..., P-1,Representing matrix ciV is arranged, the set of the row coordinate of 2 neighborhoods up and down of the element of f rows, f=0,
1 ..., P-1, P represent line number with reference to remote sensing images or remote sensing images subject to registration.
Second step：According to the following formula, using Thomas algorithm solution anisotropy row diffusion equations.
Wherein,Represent along i-th layer with reference to remote sensing images or the anisotropy metric space of remote sensing images subject to registration
Image IiV row diffusion after result, I be size and input refer to remote sensing images or remote sensing images line number subject to registration it is identical
Unit matrix, tiAnd ti+1It is the time measure value of i-th layer of anisotropy metric space and i+1 layer respectively, A2,v(Ii) table
Show the i-th tomographic image I of coded reference remote sensing images or remote sensing images anisotropy metric space subject to registrationiDiffusion coefficient square
Battle array ciThe matrix of v row,Represent i-th layer of the anisotropy metric space with reference to remote sensing images or remote sensing images subject to registration
Image IiV row, ()-1Represent inverse matrix operation.
3rd step：For v=0,1 ..., Q-1, the first step and second step in this step are repeated, is obtained Represent that edge refers to remote sensing images or remote sensing images anisotropy metric space subject to registration
The i-th tomographic image IiRow do the result after one-dimensional diffusion, Q represents the row with reference to remote sensing images or remote sensing images subject to registration
Number.
(2h) according to the following formula, calculates the i+1 with reference to remote sensing images and remote sensing images anisotropy metric space subject to registration
Tomographic image：
Wherein, Ii+1Represent the i+1 layer figure with reference to remote sensing images or remote sensing images anisotropy metric space subject to registration
Picture,Represent that the i-th tomographic image with reference to remote sensing images or remote sensing images anisotropy metric space subject to registration spreads through space
Image afterwards,Represent to pass through with reference to the i-th tomographic image of remote sensing images or remote sensing images anisotropy metric space subject to registration
Image after row diffusion.
(2i) judges whether i >=N-2 sets up, if so, terminate iteration, obtain referring to remote sensing images and remote sensing figure subject to registration
As anisotropy metric space, otherwise, i=i+1 is made, perform step (2d)；Wherein, i represents anisotropy metric space layer
Sequence number, N represent the sum with reference to remote sensing images and remote sensing images anisotropy metric space layer subject to registration.
Step 3, gradient amplitude image is calculated.
Using Sobel Operator Sobel, calculating refers to remote sensing images and remote sensing images anisotropy metric space subject to registration
Gradient amplitude figure.
According to the following formula, each tomographic image edge with reference to remote sensing images and remote sensing images anisotropy metric space subject to registration is calculated
The difference of x-axis positive direction：
Wherein, InThe n-th layer image with reference to remote sensing images or remote sensing images anisotropy metric space subject to registration is represented,Image I of the denotation coordination origin in the upper left cornernThe horizontally difference of right x-axis positive direction,Expression associative operation, n=0,
1 ..., N-1, N represent sum with reference to remote sensing images and remote sensing images anisotropy metric space layer subject to registration.
According to the following formula, each tomographic image edge with reference to remote sensing images and remote sensing images anisotropy metric space subject to registration is calculated
The difference of y-axis positive direction：
Wherein,Image I of the denotation coordination origin in the upper left cornernAlong the difference of y-axis positive direction straight down.
According to the following formula, the ladder with reference to remote sensing images and each tomographic image of remote sensing images anisotropy metric space subject to registration is calculated
Degree amplitude：
Wherein, GInRepresent the n-th layer image with reference to remote sensing images or remote sensing images anisotropy metric space subject to registration
InGradient amplitude,Represent the n-th layer image I with reference to remote sensing images or remote sensing images anisotropy metric space subject to registrationn
The horizontally difference of right x-axis positive direction,Represent empty with reference to remote sensing images or remote sensing images anisotropy yardstick subject to registration
Between n-th layer image InAlong the difference of y-axis positive direction straight down, n=0,1 ..., N-1, N represent with reference to remote sensing images and treat
The sum of registering remote sensing images anisotropy metric space layer.
Step 4, the difference of gradient amplitude image is calculated.
According to the following formula, remote sensing images and remote sensing images anisotropy subject to registration are referred to using Sobel Operator Sobel, calculating
The difference of the x-axis positive direction of the gradient amplitude image of metric space：
Wherein,Represent the gradient amplitude with reference to remote sensing images or remote sensing images anisotropy metric space subject to registration
Image GInThe horizontally difference of right x-axis positive direction, GInRepresent with reference to remote sensing images or remote sensing images subject to registration respectively to different
The n-th layer image of the gradient amplitude image of property metric space,Represent associative operation, n=0,1 ..., N-1, N represent with reference to distant
Feel the sum of image and remote sensing images anisotropy metric space layer subject to registration.
According to the following formula, remote sensing images and remote sensing images anisotropy subject to registration are referred to using Sobel Operator Sobel, calculating
The difference of the y-axis positive direction of the gradient amplitude image of metric space：
Wherein,Represent the gradient amplitude with reference to remote sensing images or remote sensing images anisotropy metric space subject to registration
Image GInAlong the difference of y-axis positive direction straight down, GInRepresent that input is each with reference to remote sensing images or remote sensing images subject to registration
The n-th layer image of the gradient amplitude image of anisotropy metric space,Represent associative operation, n=0,1 ..., N-1, N represent ginseng
Examine the sum of remote sensing images and remote sensing images anisotropy metric space layer subject to registration.
Step 5, the gradient amplitude of gradient amplitude image is calculated.
According to the following formula, the gradient amplitude of gradient amplitude image is calculated：
Wherein, GGInRepresent the gradient amplitude with reference to remote sensing images or remote sensing images anisotropy metric space subject to registration
The n-th layer image GI of imagenGradient amplitude,Expression refers to remote sensing images or remote sensing images anisotropy chi subject to registration
Spend the n-th layer image GI of the gradient amplitude image in spacenThe horizontally difference of right x-axis positive direction,Expression refers to remote sensing figure
The n-th layer image GI of the gradient amplitude image of picture or remote sensing images anisotropy metric space subject to registrationnAlong y-axis straight down
The difference of positive direction,Expression open radical sign operation, n=0,1 ..., N-1, N represent refer to remote sensing images and remote sensing figure subject to registration
As the sum of anisotropy metric space layer.
Step 6, the gradient angle of gradient amplitude image is calculated.
According to the following formula, the gradient angle of gradient amplitude image is calculated：
Wherein, AGGInRepresent the gradient width with reference to remote sensing images or remote sensing images anisotropy metric space subject to registration
Spend the n-th layer image GI of imagenGradient angle,Expression refers to remote sensing images or remote sensing images anisotropy subject to registration
The n-th layer image GI of the gradient amplitude image of metric spacenThe horizontally difference of right x-axis positive direction,Expression refers to remote sensing
The n-th layer image GI of the gradient amplitude image of image or remote sensing images anisotropy metric space subject to registrationnAlong y straight down
The difference of axle positive direction, arctan () represent the operation of four-quadrant arc tangent, n=0,1 ..., N-1, N represent to refer to remote sensing figure
The sum of picture and remote sensing images anisotropy metric space layer subject to registration.
Step 7, characteristic point is detected.
Using Harris corner detection operators, in the anisotropy yardstick with reference to remote sensing images or remote sensing images subject to registration
Characteristic point is detected on the gradient amplitude image in space.
According to the following formula, the gradient amplitude with reference to remote sensing images or remote sensing images anisotropy metric space subject to registration is calculated
The Harris matrixes of each pixel of each tomographic image of image：
Wherein, u (X, σn) represent to refer to the gradient of remote sensing images or remote sensing images anisotropy metric space subject to registration
The n-th layer image GI of magnitude imagenPosition is the Harris matrixes of X pixel, and X represents location of pixels coordinate, σnRepresent with reference to distant
Feel the scale-value of the n-th layer of image or remote sensing images anisotropy metric space subject to registration,It is x-axis direction and y-axis side
All it is to standard deviationGaussian function, * represent convolution operation.
Calculate with reference to each of the gradient amplitude image of remote sensing images or remote sensing images anisotropy metric space subject to registration
The Harris functions of each pixel of tomographic image：
R(X,σn)=det (u (X, σn))-D·tr(u(X,σn))2
Wherein, R (X, σn) represent to refer to the gradient of remote sensing images or remote sensing images anisotropy metric space subject to registration
The n-th layer image GI of magnitude imagenPosition be X pixel Harris functions, the determinant of det () expression matrix, D represent
Arbitrary constant, the summation of tr () representing matrix the elements in a main diagonal.
For R (X, σn) ＞ TthAnd R (X, σn) it is more than the point of the Harris functions at any point in its 8 neighborhood as feature
Point, obtain the spy detected on the gradient amplitude image of the anisotropy metric space with reference to remote sensing images and remote sensing images subject to registration
Levy the set of point：
Wherein, TthRepresent Harris function thresholds.
Step 8, characteristic point principal direction is generated.
By horizontal direction angle in [0,2 π] 36 deciles.
It is 6 σ by the radius of set of characteristic pointsnCircle shaped neighborhood region region in pixel gradient direction AGGIn(X) it is, it is determined that circular
Decile angular range in region where pixel, wherein, σnRepresent with reference to remote sensing images or remote sensing images subject to registration respectively to different
The scale-value of the n-th layer of property metric space, AGGIn(X) represent that position coordinates is X pixel in circle shaped neighborhood region around characteristic point
Gradient angle, n=0,1 ..., N-1, N represent with reference to remote sensing images and remote sensing images anisotropy metric space layer subject to registration
Sum.
The gradient magnitude GGI of all pixels to add up respectively in the range of each decile anglen(X) gradient, is formed
Direction histogram, wherein, GGIn(X) represent that position coordinates is the gradient amplitude of X pixel, n in circle shaped neighborhood region around characteristic point
=0,1 ..., N-1, N represent sum with reference to remote sensing images and remote sensing images anisotropy metric space layer subject to registration.
Gradient direction corresponding to the numerical value of 0.8 times of maximum, the master as characteristic point will be more than in gradient orientation histogram
Direction.
Step 9, feature point description is generated.
The circle shaped neighborhood region rotate counterclockwise θ of (9a) by radius for ρ characteristic pointmDegree, is obtained identical shown in accompanying drawing 2 (a)
ρ points are 3 sections along radial direction by the circle shaped neighborhood region of radius, circle shaped neighborhood region, and the inner circle radius of neighbourhood is 0.25 ρ, and middle circle is adjacent
Domain radius is 0.73 ρ, and the cylindrical radius of neighbourhood is ρ, and [0,2 π] is divided into 8 sections by circle shaped neighborhood region along angle direction, and inner circle is made
For an entirety, circle shaped neighborhood region is divided the subregion of the area equation for 17 diverse locations around characteristic point, wherein, ρ
Value be 12 σn, σnReference remote sensing images or remote sensing images anisotropy metric space subject to registration where expression characteristic point
N-th layer scale-value, n=0,1 ..., N-1, N represent refer to remote sensing images and remote sensing images anisotropy yardstick subject to registration
The sum of space layer, θmRepresent the principal direction of this feature point.
The cartesian coordinate of pixel in characteristic point circle shaped neighborhood region is converted to log-polar, log-polar angle by (9b)
Direction level to the right, in accompanying drawing 2 (a) in the Arabic numerals region respective figure 2 (b) of mark mark Arabic numerals area
Domain, is divided into 8 sections in the range of [0,2 π], log-polar logarithm length direction straight down,Scope
It is interior it is non-be divided into 3 sections, wherein, ρ represent characteristic point around circle shaped neighborhood region radius.
All pixels in (9c) log-polar grid in every sub-regions are according to its gradient amplitude GGInAnd gradient (X)
Direction AGGIn(X) gradient orientation histogram is calculated, each sub-regions form the gradient direction vector of one 8 dimension, spelled successively
The description for the characteristic point that the gradient direction vector for connecing 17 subregions is formed one 136 dimension is sub, wherein, gradient orientation histogram
Angle 8 sections, GGI are divided into the range of [0,2 π]n(X) represent that position coordinates is in circle shaped neighborhood region around characteristic point
The gradient amplitude of X pixel, AGGIn(X) represent that position coordinates is the gradient angle of X pixel in circle shaped neighborhood region around characteristic point
Degree.
(9d) is using step (9a), step (9b), the same procedure of step (9c), generation set of characteristic points CIRDescription son
Set DIRWith set of characteristic points CISDescription subclass DIS,Its
In, DIRRepresent description of characteristic point detected on the gradient amplitude image with reference to the anisotropy metric space of remote sensing images
Set, DISThe characteristic point for representing to detect on the gradient amplitude image of the anisotropy metric space of remote sensing images subject to registration is retouched
State subclass, L1The characteristic point that expression detects on the gradient amplitude image with reference to the anisotropy metric space of remote sensing images
Description son sum, L2Represent the feature detected on the gradient amplitude image of the anisotropy metric space of remote sensing images subject to registration
The description son sum of point.
Step 10, Feature Points Matching.
Match on the anisotropy metric space gradient amplitude image with reference to remote sensing images the characteristic point that detects and
The characteristic point detected on the anisotropy metric space gradient amplitude image of remote sensing images subject to registration：
The sub- sequence number k of description of Characteristics of The Remote Sensing Images point subject to registration is initialized as 1 by (10a).
(10b) according to the following formula, calculates description of Characteristics of The Remote Sensing Images point subject to registrationWith with reference to remote sensing images each
The vector of the Euclidean distance composition of description of characteristic point：
Wherein, dkRepresent description of Characteristics of The Remote Sensing Images point subject to registrationWith each description with reference to remote sensing images
Euclidean distance composition vector,dk,jRepresent description of registering Characteristics of The Remote Sensing Images pointWith reference
Description of Characteristics of The Remote Sensing Images pointEuclidean distance,K-th of description of Characteristics of The Remote Sensing Images point subject to registration is represented,Represent j-th of description with reference to Characteristics of The Remote Sensing Images point, j=1,2 ..., L1, L1Expression refers to Characteristics of The Remote Sensing Images point
The number of son is described, | | | | expression takes norm to operate.
(10c) is to vectorial dkIn L1Individual element sorts to obtain vectorial s from small to largek。
(10d) ifThen descriptionCorresponding characteristic point and with description sonEuclidean distance be sk,0's
The Feature Points Matching with reference to corresponding to description of remote sensing images；IfAndThen descriptionCorresponding spy
Levy point and with describing sonEuclidean distance be dkPreceding E element reference remote sensing images description corresponding to characteristic point
Match somebody with somebody；IfThen descriptionAny Feature Points Matching of the corresponding characteristic point discord with reference to remote sensing images；
Wherein, sk,0Represent vectorial skFirst element, sk,1Represent vectorial skSecond element, TLAnd THRepresent most
Neighbour and time nearest neighbor distance ratio threshold value, E span are the integer in [2,3].
(10e) judges k >=L2Whether set up, if so, obtain remote sensing images subject to registration and the feature with reference to remote sensing images
Point matching relation CIR,IS,Perform step (11)；Otherwise, k=k+1 is made, performs this step
The step of (10b)；
Wherein, k represents the sub- sequence number of description of Characteristics of The Remote Sensing Images point subject to registration, CIR,ISRepresent with reference to remote sensing images it is each to
The characteristic point detected on the gradient amplitude image of different in nature metric space and the anisotropy yardstick sky in remote sensing images subject to registration
Between gradient amplitude image on the pair relationhip set of characteristic point that detects, L3Represent the point pair of Feature Points Matching stage match
Sum.
Step 11, the characteristic point pair of deletion error matching.
Referring to the drawings 3, using the consistent RANSAC algorithms of random sampling, the matching double points of deletion error, obtain subject to registration distant
Image is felt to the geometric transform relation with reference to remote sensing images, is comprised the following steps that：
(11a) loop initialization times N um is 0, and it is 0 to initialize the matching double points number included in optimal consistent point set C.
(11b) Num=Num+1, judges Num>Whether 1000 set up, if so, perform step (11i)；Otherwise, step is performed
Suddenly (11c).
(11c) is from Feature Points Matching relation CIR,ISIn randomly select 3 different matching characteristic points pair；
Wherein, CIR,ISRepresent the spy detected on the gradient amplitude image with reference to the anisotropy metric space of remote sensing images
The pairing of sign point and the characteristic point detected on the gradient amplitude image of the anisotropy metric space of remote sensing images subject to registration
Set of relationship.
(11d) uses affine Transform Model, calculates the transformation matrix T1 for meeting 3 different matching characteristic points pair.
(11e) calculates Feature Points Matching relation CIR,ISThe middle consistent point set Con for meeting transformation matrix T1；
Wherein, CIR,ISRepresent the spy detected on the gradient amplitude image with reference to the anisotropy metric space of remote sensing images
The pairing of sign point and the characteristic point detected on the gradient amplitude image of the anisotropy metric space of remote sensing images subject to registration
Set of relationship.
The point that (11f) judges to include in consistent point set Con is more than whether consistent point is set up to threshold value TH to number, if into
It is vertical, perform step (11g)；Otherwise, step (11b) is performed；
Wherein, TH represents consistent point to threshold value, and TH values are L30.1 times, L3Represent the point of Feature Points Matching stage match
To sum.
The point that (11g) judges to include in consistent point set Con is more than whether current optimal consistent point is set up to number to number,
If so, perform step (11h)；Otherwise, step (11b) is performed.
(11h) consistent point set Con replaces current optimal consistent point set C, and using affine Transform Model, calculating meets consistent point
Collect C transformation matrix T2, perform step (11b).
The current optimal consistent point set C and transformation matrix T2 of (11i) output.
The characteristic point correctly matched with reference to remote sensing images and remote sensing images subject to registration is included in the optimal consistent point set C of output
It is right；Transformation matrix T2 be obtain remote sensing images subject to registration to refer to remote sensing images geometric transform relation T.
Step 12, registering reference picture and image subject to registration.
According to remote sensing images subject to registration to the geometric transform relation T with reference to remote sensing images, matched somebody with somebody using bilinear interpolation method
Standard refers to remote sensing images and remote sensing images subject to registration.
The effect of the present invention is described further with reference to analogous diagram.
1. simulated conditions：
Emulation experiment of the present invention under the systems of 380 2.53GHz Windows of Intel (R) Core (TM) i3 CPU M 7,
On Matlab 2010a operation platforms, the emulation experiment of the present invention is completed.
2. emulation experiment content：
It is 3 to be taken in the emulation experiment of the present invention with reference to the group number G of remote sensing and remote sensing images metric space subject to registration, organizes internal layer
Number L is 3, and therefore, the total N of the layer of anisotropy metric space is 9；Station meter angle value σ0For 1.6, ginseng is inputted in emulation experiment
Examine remote sensing images and remote sensing image data subject to registration is converted to floating type between 0-1, Harris function thresholds Tth
For 0.005, threshold value T in Feature Points MatchingLAnd TH0.6 and 0.97 are respectively set to, geometric transformation model is affine change between image
Change.The test remote sensing images of the emulation experiment input of the present invention are divided into two classes：(1) multi-spectral remote sensing image pair, (2) multi-source remote sensing
Image pair.
Accompanying drawing 4 (a) and accompanying drawing 4 (b) be first group of multi-spectral remote sensing image to P-A, wherein, accompanying drawing 4 (a) is reference chart
Picture, size are 761 × 748 pixels, and sensor type is Landsat-7ETM+, and 5 wave bands, it is 2000/7/24 to obtain the time；It is attached
Fig. 4 (b) is image subject to registration, and size is 761 × 748 pixels, and sensor type is the -5TM of Landsat 4,3 wave bands, during acquisition
Between be 1999/7/6.
Accompanying drawing 5 (a) and accompanying drawing 5 (b) be second group of multi-spectral remote sensing image to P-B, wherein, accompanying drawing 5 (a) is reference chart
Picture, size are 761 × 748 pixels, and sensor type is Landsat-7ETM+, and 3 wave bands, it is 2003/4/12 to obtain the time；It is attached
Fig. 5 (b) is image subject to registration, and size is 761 × 748 pixels, and sensor type is the -5TM of Landsat 4,5 wave bands, during acquisition
Between be 2006/06/15.
Accompanying drawing 6 (a) and accompanying drawing 6 (b) be multi-source Remote Sensing Images to P-C, wherein, accompanying drawing 6 (a) is reference picture, and size is
800 × 800 pixels, sensor type are ALOS-PALSAR, and it is 2010/6/5 to obtain the time；Accompanying drawing 6 (b) is image subject to registration,
Size is 800 × 800 pixels, and sensor type is Landsat ETM+, and it is 1999/6/26 to obtain the time.
The SIFT algorithms of classics are contrasted in the emulation experiment of the present invention first, it is nearest to also provide a comparison the dual threshold proposed in text
Adjacent matching criterior and classical single threshold arest neighbors matching criterior, therefore, different to two kinds of application for each test image
The method proposed in method and text is contrasted, and both approaches are SIFT algorithms and Harris anisotropic gradient metric spaces
+ nearest neighbor distance is briefly referred to as SIFT and HNNR than matching+RANSAC algorithms (Harris-NDGSS-NNDR-RANSAC),
The complete algorithm mentioned in text is Harris anisotropic gradients metric space+dual threshold nearest neighbor distance than matching+RANSAC
Algorithm (Harris-NDGSS-DNNDR-RANSAC), referred to as HNDR.Make 3 kinds of different methods pair in emulation experiment of the present invention
Same image to producing same amount of characteristic point as far as possible, shown in the simulation experiment result following table.
3. matching performance appraisal procedure：
Using the characteristic point correctly matched to number NcWith root-mean-square error RMSE as the standard for assessing registering performance.Nc
It is the characteristic point using the correct matching obtained after the consistent RANSAC algorithms of random sampling to number.
Root-mean-square error RMSE is calculated according to following formula：
Wherein, xrAnd yrRepresent respectively with reference to r-th point of the row coordinate and row coordinate chosen manually on remote sensing images,
WithRepresent to choose manually on remote sensing images subject to registration respectively r-th point is by remote sensing images subject to registration to referring to remote sensing images
Geometric transform relation T conversion after row coordinate and row coordinate, r=1,2 ..., NGT, NGTRepresent from reference remote sensing images and treat
The true match point chosen manually on registering remote sensing images is to sum.
4. the simulation experiment result and analysis：
From annex Fig. 4 (a) with Fig. 4 (b) as can be seen that identical with the major part of remote sensing images subject to registration with reference to remote sensing images
Regional luminance inverts；As shown in the table, SIFT algorithms can not realize correct registration, but be based on anisotropic gradient metric space
HNNR and HNDR algorithms can detect more correct matching double points, demonstrate article proposition algorithm can registering brightness of image it is anti-
The remote sensing images pair turned.Accompanying drawing 5 (a) and accompanying drawing 5 (b) and multi-spectral remote sensing image pair, two width figure respective regions brightness of image
Reversion or nonlinear change；As shown in the table, the method based on SIFT also can accurately realize registration, but be based on anisotropy
HNNR the and HNDR algorithms in gradient scale space can detect more correct matching double points, also demonstrate the algorithm energy of article proposition
Registering brightness of image nonlinear change or the remote sensing images pair of brightness reversion.Accompanying drawing 6 (a) and accompanying drawing 6 (b) are multi-source remote sensing figures
As right, linear change pattern is mainly presented in same area brightness of image, as shown in the table, and 3 kinds of methods can realize correct registration,
And SIFT algorithms can also realize more correct matching pair, because SIFT algorithms predominantly detect image texture structure, in line
More valuable point features can be extracted on the reason preferable image of Edge preservation, also demonstrates and matches somebody with somebody in a range of remote sensing images
SIFT algorithms can be used in quasi- task.Wherein, accompanying drawing 4 (c), accompanying drawing 5 (c) and accompanying drawing 6 (c) are remote sensing images subject to registration to ginseng
Examine the registration result after remote sensing images coordinate system transformation, accompanying drawing 4 (d), accompanying drawing 5 (d) and accompanying drawing 6 (d) be with reference to remote sensing images and
The fusion of remote sensing images subject to registration after conversion.
It is as shown in the table to the registration result of progress to remote sensing images using HNNR algorithms and HNDR algorithms.Can by following table
To find out, the feature point number that HNDR algorithms correctly match all is more than the feature point number that HNNR algorithms correctly match, and HNDR is calculated
The dual threshold nearest neighbor distance ratio matching that method and the difference of HNNR algorithms propose in HNDR algorithms using text is accurate
Then carry out Feature Points Matching, and HNNR algorithms using the nearest neighbor distance of classics than matching criterior, thus demonstrate double
Threshold value nearest neighbor distance significantly more efficient can carry out Feature Points Matching than matching process using feature point description sub-information.Pass through ratio
Compared with root-mean-square error RMSE, it can be seen that the method for proposition can realize the registration of subpixel accuracy.In a word, HNDR algorithms are not only
The registration task of multi-source Remote Sensing Images can preferably be completed, at the same overcome existing algorithm registration brightness of image exist it is larger non-thread
Property change even brightness of image reversion multi-spectral remote sensing image it is low to accuracy the problem of.
Wherein, Pair represents the remote sensing images pair of test, and Method represents the distinct methods of matching remote sensing images pair, NIRWith
NISThe feature point number detected with reference to remote sensing images and remote sensing images subject to registration, N are represented respectivelycRepresent the feature correctly matched
Point represents root-mean-square error to number, RMSE.
Claims (5)
1. a kind of remote sensing image registration method based on anisotropic gradient metric space, comprises the following steps：
(1) input refers to remote sensing images and remote sensing images subject to registration；
(2) metric space of structural anisotropy's diffusion：
(2a) calculates the scale-value of each layer of anisotropy metric space with reference to remote sensing images and remote sensing images subject to registration；
Scale-value is transformed into time measure value by (2b)；
The reference remote sensing images and remote sensing images subject to registration of (2c) to input, use standard deviation as σ0Gaussian filtering, referred to
Remote sensing images and the tomographic image of remote sensing images anisotropy metric space the 0th subject to registration；
The sequence number i of anisotropy metric space layer is initialized as zero by (2d)；
(2e) according to the following formula, calculates i-th layer with reference to remote sensing images and remote sensing images anisotropy metric space subject to registration respectively
The diffusion coefficient matrix of image：
Wherein, ciRepresent the diffusion of the i-th tomographic image with reference to remote sensing images or remote sensing images anisotropy metric space subject to registration
Coefficient matrix,Represent the i-th tomographic image I with reference to remote sensing images or remote sensing images anisotropy metric space subject to registrationiUsing
Standard deviation is the image after 1 gaussian filtering,Represent the image after gaussian filteringGradient amplitude, | | represent modulus
Operation, K represent contrast factor, and K value is gradient amplitudeThe hundredths of statistic histogram 70% gradient amplitude
Value；
(2f) according to the following formula, is done to the i-th tomographic image with reference to remote sensing images and remote sensing images anisotropy metric space subject to registration
Row diffusion：
Wherein,Represent along the i-th tomographic image I with reference to remote sensing images or remote sensing images anisotropy metric space subject to registrationi
Row diffusion after image, I1Represent that line number and columns are identical with the columns with reference to remote sensing images or remote sensing images subject to registration
Unit square formation, tiAnd ti+1I-th with reference to remote sensing images or remote sensing images anisotropy metric space subject to registration is represented respectively
The time measure value of layer and i+1 layer, A (Ii) presentation code refers to remote sensing images or remote sensing images anisotropy chi subject to registration
Spend the i-th tomographic image diffusion coefficient c in spaceiMatrix, IiExpression refers to remote sensing images or remote sensing images anisotropy subject to registration
I-th tomographic image of metric space, ()-1Represent inverse matrix operation；
(2g) according to the following formula, is done to the i-th tomographic image with reference to remote sensing images and remote sensing images anisotropy metric space subject to registration
Row diffusion：
Wherein,Represent along the i-th tomographic image I with reference to remote sensing images or remote sensing images anisotropy metric space subject to registrationi
Row diffusion after image, I2Represent that line number and columns are identical with the line number with reference to remote sensing images or remote sensing images subject to registration
Unit square formation；
(2h) according to the following formula, calculates the i+1 layer figure with reference to remote sensing images and remote sensing images anisotropy metric space subject to registration
Picture：
Wherein, Ii+1Represent the i+1 tomographic image with reference to remote sensing images or remote sensing images anisotropy metric space subject to registration；
(2i) judges whether i >=N-2 sets up, if so, obtain the anisotropy with reference to remote sensing images and remote sensing images subject to registration
Metric space, otherwise, i=i+1 is made, perform step (2e)；Wherein, N represents each with reference to remote sensing images and remote sensing images subject to registration
The sum of anisotropy metric space layer；
(3) gradient amplitude image is calculated：
Using Sobel Operator Sobel, the ladder with reference to remote sensing images and remote sensing images anisotropy metric space subject to registration is calculated
Spend magnitude image；
(4) difference of gradient amplitude image is calculated：
(4a) according to the following formula, using Sobel Operator Sobel, calculating refers to remote sensing images and remote sensing images anisotropy subject to registration
The difference of the x-axis positive direction of the gradient amplitude image of metric space：
Wherein,Represent the gradient amplitude image with reference to remote sensing images or remote sensing images anisotropy metric space subject to registration
GInThe horizontally difference of right x-axis positive direction, GInExpression refers to remote sensing images or remote sensing images anisotropy chi subject to registration
The n-th layer image of the gradient amplitude image in space is spent,Represent associative operation；
(4b) according to the following formula, using Sobel Operator Sobel, calculating refers to remote sensing images and remote sensing images anisotropy subject to registration
The difference of the y-axis positive direction of the gradient amplitude image of metric space：
Wherein,Represent the gradient amplitude image with reference to remote sensing images or remote sensing images anisotropy metric space subject to registration
GInAlong the difference of y-axis positive direction straight down；
(5) according to the following formula, the gradient amplitude of gradient amplitude image is calculated：
Wherein, GGInRepresent the gradient amplitude image with reference to remote sensing images or remote sensing images anisotropy metric space subject to registration
N-th layer image GInGradient amplitude,Radical sign operation is opened in expression；
(6) according to the following formula, the gradient angle of gradient amplitude image is calculated：
Wherein, AGGInRepresent the gradient amplitude image with reference to remote sensing images or remote sensing images anisotropy metric space subject to registration
N-th layer image GInGradient angle, arctan () represents the operation of four-quadrant arc tangent；
(7) characteristic point is detected：
Using Harris corner detection operators, in the anisotropy metric space with reference to remote sensing images and remote sensing images subject to registration
Characteristic point is detected on gradient amplitude image, obtains set of characteristic points：
Wherein, CIRRepresent the feature point set detected on the gradient amplitude image with reference to the anisotropy metric space of remote sensing images
Close, CISRepresent the set of characteristic points detected on the gradient amplitude image of the anisotropy metric space of remote sensing images subject to registration, R
The sum of characteristic point detected on the gradient amplitude image with reference to the anisotropy metric space of remote sensing images is represented, S is represented
The sum of the characteristic point detected on the gradient amplitude image of the anisotropy metric space of remote sensing images subject to registration；
(8) characteristic point principal direction is generated：
(8a) 36 deciles in [0,2 π] by horizontal direction angle；
(8b) is 6 σ by the radius of set of characteristic pointsnCircle shaped neighborhood region region in pixel gradient direction AGGIn(X) it is, it is determined that circular
Decile angular range in region where pixel, wherein, σnRepresent with reference to remote sensing images or remote sensing images subject to registration respectively to different
The scale-value of the n-th layer of property metric space, AGGIn(X) represent that position coordinates is X pixel in circle shaped neighborhood region around characteristic point
Gradient angle；
The gradient magnitude GGI for all pixels that (8c) adds up in the range of each decile angle respectivelyn(X) gradient side, is formed
To histogram, wherein, GGIn(X) represent that position coordinates is the gradient amplitude of X pixel in circle shaped neighborhood region around characteristic point；
(8d) will be more than gradient direction corresponding to the numerical value of 0.8 times of maximum, the master as characteristic point in gradient orientation histogram
Direction；
(9) feature point description is generated：
The circle shaped neighborhood region rotate counterclockwise θ of (9a) by radius for ρ characteristic pointmρ points are 3 along radial direction by degree, circle shaped neighborhood region
Section, the inner circle radius of neighbourhood are 0.25 ρ, and the middle circle radius of neighbourhood is 0.73 ρ, and the cylindrical radius of neighbourhood is ρ, and circle shaped neighborhood region is along angle
[0,2 π] is divided into 8 sections by direction, and inner circle is as an entirety, and circle shaped neighborhood region is divided for 17 not around characteristic point
With the subregion of the area equation of position, wherein, ρ value is 12 σn, θmRepresent the principal direction of this feature point；
The cartesian coordinate of pixel in characteristic point circle shaped neighborhood region is converted to log-polar, log-polar angle direction by (9b)
It is horizontal to be divided into 8 sections in the range of [0,2 π] to the right, log-polar logarithm length direction straight down,In the range of it is non-be divided into 3 sections, wherein, ρ represent characteristic point around circle shaped neighborhood region radius；
All pixels in (9c) log-polar grid in every sub-regions are according to its gradient amplitude GGInAnd gradient direction (X)
AGGIn(X) gradient orientation histogram is calculated, each sub-regions form the gradient direction vector of one 8 dimension, splice 17 successively
The gradient direction vector of subregion is formed description of the characteristic point of one 136 dimension, wherein, the angle of gradient orientation histogram
Degree is divided into 8 sections in the range of [0,2 π]；
(9d) is using step (9a), step (9b), the same procedure of step (9c), generation set of characteristic points CIRDescription subclass
DIRWith set of characteristic points CISDescription subclass DIS,Wherein,
DIRRepresent the description subset of characteristic point detected on the gradient amplitude image with reference to the anisotropy metric space of remote sensing images
Close, DISRepresent the description of characteristic point detected on the gradient amplitude image of the anisotropy metric space of remote sensing images subject to registration
Subclass, L1The characteristic point for representing to detect on the gradient amplitude image with reference to the anisotropy metric space of remote sensing images is retouched
State sub- sum, L2Represent the characteristic point detected on the gradient amplitude image of the anisotropy metric space of remote sensing images subject to registration
Description son sum；
(10) Feature Points Matching：
The sub- sequence number k of description of Characteristics of The Remote Sensing Images point subject to registration is initialized as 1 by (10a)；
(10b) according to the following formula, calculates description of Characteristics of The Remote Sensing Images point subject to registrationWith each feature of reference remote sensing images
The vector of the Euclidean distance composition of description of point：
Wherein, dkRepresent description of Characteristics of The Remote Sensing Images point subject to registrationWith retouching for each characteristic point with reference to remote sensing images
The vector of the Euclidean distance composition of son is stated,dk,jRepresent description of registering Characteristics of The Remote Sensing Images pointWith
With reference to description of Characteristics of The Remote Sensing Images pointEuclidean distance,Represent Characteristics of The Remote Sensing Images point subject to registration k-th is retouched
State son,Represent j-th of description with reference to Characteristics of The Remote Sensing Images point, j=1,2 ..., L1, L1Represent special with reference to remote sensing images
The number of sign point description, | | | | expression takes norm to operate；
(10c) is to vectorial dkIn L1Individual element sorts to obtain vectorial s from small to largek；
(10d) ifThen descriptionCorresponding characteristic point and with description sonEuclidean distance be sk,0Reference
Feature Points Matching corresponding to description of remote sensing images；IfAndThen descriptionCorresponding characteristic point
With with description sonEuclidean distance be dkPreceding E element reference remote sensing images description corresponding to Feature Points Matching；
IfThen descriptionAny Feature Points Matching of the corresponding characteristic point discord with reference to remote sensing images；
Wherein, sk,0Represent vectorial skFirst element, sk,1Represent vectorial skSecond element, TLAnd THRepresent arest neighbors and
For secondary nearest neighbor distance than threshold value, E span is the integer in [2,3]；
(10e) judges k >=L2Whether set up, if so, obtain remote sensing images subject to registration and the Feature Points Matching with reference to remote sensing images
Relation CIR,IS,Perform step (11)；Otherwise, k=k+1 is made, performs step (10b)；
Wherein, CIR,ISRepresent the characteristic point detected on the gradient amplitude image with reference to the anisotropy metric space of remote sensing images
With the pair relationhip of characteristic point detected on the gradient amplitude image of the anisotropy metric space of remote sensing images subject to registration
Set, L3Represent the point of Feature Points Matching stage match to sum；
(11) characteristic point pair of deletion error matching：
Using the consistent RANSAC algorithms of random sampling, the matching double points of deletion error, remote sensing images subject to registration are obtained to reference to distant
Feel the geometric transform relation of image；
(12) registering reference picture and image subject to registration：
According to geometric transform relation, remote sensing images and remote sensing images subject to registration are referred to using bilinear interpolation method registration.
2. the remote sensing image registration method according to claim 1 based on anisotropic gradient metric space, its feature exist
In the calculating described in step (2a) refers to the yardstick of each layer of anisotropy metric space of remote sensing images and remote sensing images subject to registration
Value obtains according to the following formula：
σe(m, u)=σ02m+u/L
Wherein, σeThe scale-value of the e layers with reference to remote sensing images or remote sensing images anisotropy metric space subject to registration is represented,
σ0The station meter angle value with reference to remote sensing images or remote sensing images subject to registration is represented, by different σeThe anisotropy yardstick of composition is empty
Between common G groups, every group of L layer, m represents the group index of anisotropy metric space, m=0,1 ..., G-1, u represent every group of internal layer
Index, u=0,1 ..., L-1, e=0,1,2 ..., N-1.
3. the remote sensing image registration method according to claim 1 based on anisotropic gradient metric space, its feature exist
In the time measure value that is transformed into scale-value described in step (2b) is carried out according to the following formula：
Wherein, teRepresent the time measure of the e layers with reference to remote sensing images or remote sensing images anisotropy metric space subject to registration
Value, σeRepresent the scale-value of the e layers with reference to remote sensing images or remote sensing images anisotropy metric space subject to registration, e=0,
1,2 ..., N-1, m represent the group index of anisotropy metric space, m=0,1 ..., G-1, G represent with reference to remote sensing images and
The sum of remote sensing images anisotropy metric space group subject to registration, u represent the index of every group of internal layer, u=0,1 ..., L-1, L table
Show the sum with reference to remote sensing images and every group of layer of remote sensing images anisotropy metric space subject to registration.
4. the remote sensing image registration method according to claim 1 based on anisotropic gradient metric space, its feature exist
In the use Sobel Operator Sobel described in step (3), calculating is with reference to remote sensing images and remote sensing images subject to registration respectively to different
The step of gradient amplitude image of property metric space, is as follows：
The first step：According to the following formula, each layer figure with reference to remote sensing images and remote sensing images anisotropy metric space subject to registration is calculated
As the difference along x-axis positive direction：
Wherein, InThe n-th layer image with reference to remote sensing images or remote sensing images anisotropy metric space subject to registration is represented,Table
Show image I of the origin of coordinates in the upper left cornernThe horizontally difference of right x-axis positive direction,Represent associative operation, n=0,1 ...,
N-1, N represent the sum with reference to remote sensing images and remote sensing images anisotropy metric space layer subject to registration；
Second step：According to the following formula, each layer figure with reference to remote sensing images and remote sensing images anisotropy metric space subject to registration is calculated
As the difference along y-axis positive direction：
Wherein,Image I of the denotation coordination origin in the upper left cornernAlong the difference of y-axis positive direction straight down；
3rd step：According to the following formula, calculate and refer to remote sensing images and each tomographic image of remote sensing images anisotropy metric space subject to registration
Gradient amplitude：
Wherein, GInRepresent the n-th layer image I with reference to remote sensing images or remote sensing images anisotropy metric space subject to registrationn's
Gradient amplitude.
5. the remote sensing image registration method according to claim 1 based on anisotropic gradient metric space, its feature exist
In, use Harris corner detection operators described in step (7), with reference to remote sensing images and remote sensing images subject to registration it is each to
Characteristic point is detected on the gradient amplitude image of different in nature metric space to carry out in accordance with the following steps：
The first step：According to the following formula, the gradient width with reference to remote sensing images and remote sensing images anisotropy metric space subject to registration is calculated
Spend the Harris matrixes of each pixel of each tomographic image of image：
Wherein, u (X, σn) represent to refer to the gradient amplitude figure of remote sensing images or remote sensing images anisotropy metric space subject to registration
The n-th layer image GI of picturenPosition is the Harris matrixes of X pixel, and X represents location of pixels coordinate,It is x-axis direction and y
Direction of principal axis standard deviation is allGaussian function, * represent convolution operation；
Second step：Calculate with reference to each of the gradient amplitude image of remote sensing images and remote sensing images anisotropy metric space subject to registration
The Harris functions of each pixel of tomographic image：
R(X,σn)=det (u (X, σn))-D·tr(u(X,σn))2
Wherein, R (X, σn) represent to refer to the gradient amplitude figure of remote sensing images or remote sensing images anisotropy metric space subject to registration
The n-th layer image GI of picturenPosition is the Harris functions of X pixel, the determinant of det () expression matrix, and D represents any normal
Number, the summation of tr () representing matrix the elements in a main diagonal；
3rd step：For R (X, σn) ＞ TthAnd R (X, σn) it is more than the point of the Harris functions at any point in its 8 neighborhood as special
Point is levied, obtains what is detected on the gradient amplitude image of the anisotropy metric space with reference to remote sensing images and remote sensing images subject to registration
The set of characteristic point：
Wherein, TthRepresent Harris function thresholds.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510770880.5A CN105427298B (en) | 2015-11-12 | 2015-11-12 | Remote sensing image registration method based on anisotropic gradient metric space |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510770880.5A CN105427298B (en) | 2015-11-12 | 2015-11-12 | Remote sensing image registration method based on anisotropic gradient metric space |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105427298A CN105427298A (en) | 2016-03-23 |
| CN105427298B true CN105427298B (en) | 2018-03-06 |
Family
ID=55505479
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510770880.5A Active CN105427298B (en) | 2015-11-12 | 2015-11-12 | Remote sensing image registration method based on anisotropic gradient metric space |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105427298B (en) |
Families Citing this family (20)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106023183B (en) * | 2016-05-16 | 2019-01-11 | 西北工业大学 | A kind of real-time Algorism of Matching Line Segments method |
| WO2018076137A1 (en) * | 2016-10-24 | 2018-05-03 | 深圳大学 | Method and device for obtaining hyper-spectral image feature descriptor |
| CN109003293A (en) * | 2017-06-07 | 2018-12-14 | 北京航空航天大学 | Inhibit the SAR image registration method of model based on anisotropy spot |
| CN107563438B (en) | 2017-08-31 | 2019-08-30 | 西南交通大学 | A kind of multi-modal Remote Sensing Images Matching Method and system of fast robust |
| CN107657597B (en) * | 2017-10-19 | 2020-09-08 | 中国科学院遥感与数字地球研究所 | Automatic geometric correction method for cross-platform moon-based earth observation image |
| WO2019062166A1 (en) | 2017-09-30 | 2019-04-04 | 中国科学院遥感与数字地球研究所 | Method for automatic geometric correction of cross-platform moon-based earth observation imaging |
| CN107909018B (en) * | 2017-11-06 | 2019-12-06 | 西南交通大学 | Stable multi-mode remote sensing image matching method and system |
| CN108009595B (en) * | 2017-12-25 | 2018-10-12 | 北京航空航天大学 | A kind of image-recognizing method of feature based stipulations |
| CN108346162B (en) * | 2018-03-26 | 2019-10-11 | 西安电子科技大学 | Remote sensing image registration method based on structural information and space constraint |
| CN110464379B (en) * | 2018-05-11 | 2022-10-11 | 深圳市理邦精密仪器股份有限公司 | Fetal head circumference measuring method and device and terminal equipment |
| CN110501728B (en) * | 2018-05-16 | 2022-03-29 | 清华大学 | Frequency discrimination method and device for time hopping signal of positioning base station |
| CN110096540B (en) * | 2019-04-16 | 2022-02-18 | 湖北地信科技集团股份有限公司 | Mapping data conversion method, device, storage medium and device |
| CN110458876B (en) * | 2019-08-08 | 2023-01-31 | 哈尔滨工业大学 | Multi-temporal POLSAR image registration method based on SAR-SIFT features |
| CN111028201B (en) * | 2019-11-13 | 2023-10-27 | 东北大学 | Fundus blood vessel image segmentation system and method based on multi-scale level set |
| CN111125414B (en) * | 2019-12-26 | 2023-08-18 | 郑州航空工业管理学院 | Automatic searching method for specific target of unmanned aerial vehicle remote sensing image |
| CN111223133B (en) * | 2020-01-07 | 2022-10-11 | 上海交通大学 | Registration method of heterogeneous images |
| CN112784761A (en) * | 2021-01-26 | 2021-05-11 | 哈尔滨理工大学 | Special texture background remote sensing image matching method |
| CN113240743B (en) * | 2021-05-18 | 2022-03-25 | 浙江大学 | Heterogeneous image pose estimation and registration method, device and medium based on neural network |
| CN113379808B (en) * | 2021-06-21 | 2022-08-12 | 昆明理工大学 | Method for registration of multiband solar images |
| CN114820739B (en) * | 2022-07-01 | 2022-10-11 | 浙江工商大学 | Multispectral camera-oriented image rapid registration method and device |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103198483A (en) * | 2013-04-07 | 2013-07-10 | 西安电子科技大学 | Multiple time phase remote sensing image registration method based on edge and spectral reflectivity curve |
| CN104794678A (en) * | 2015-05-04 | 2015-07-22 | 福建师范大学 | Automatic registration method for high-spatial-resolution remote-sensing images based on SIFI feature points |
-
2015
- 2015-11-12 CN CN201510770880.5A patent/CN105427298B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103198483A (en) * | 2013-04-07 | 2013-07-10 | 西安电子科技大学 | Multiple time phase remote sensing image registration method based on edge and spectral reflectivity curve |
| CN104794678A (en) * | 2015-05-04 | 2015-07-22 | 福建师范大学 | Automatic registration method for high-spatial-resolution remote-sensing images based on SIFI feature points |
Non-Patent Citations (2)
| Title |
|---|
| IR remote sensing image registration based on multi-scale feature extraction;Jun Kong等;《2014 International Joint Conference on Neural Networks》;20140711;第1352-1358页 * |
| 基于梯度尺度空间的遥感影像多尺度分割方法及应用研究;张万强等;《安徽农业科学》;20131231;第41卷(第30期);第12190-12195页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105427298A (en) | 2016-03-23 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105427298B (en) | Remote sensing image registration method based on anisotropic gradient metric space | |
| Ye et al. | Fast and robust matching for multimodal remote sensing image registration | |
| Zuo et al. | A robust approach to reading recognition of pointer meters based on improved mask-RCNN | |
| CN104574347B (en) | Satellite in orbit image geometry positioning accuracy evaluation method based on multi- source Remote Sensing Data data | |
| CN104200461B (en) | The remote sensing image registration method of block and sift features is selected based on mutual information image | |
| CN111833237B (en) | Image registration method based on convolutional neural network and local homography transformation | |
| CN106991695A (en) | A kind of method for registering images and device | |
| CN106023187B (en) | A kind of method for registering images based on SIFT feature and angle relative distance | |
| CN106919944A (en) | A kind of wide-angle image method for quickly identifying based on ORB algorithms | |
| CN107993258A (en) | A kind of method for registering images and device | |
| JP5289412B2 (en) | Local feature amount calculation apparatus and method, and corresponding point search apparatus and method | |
| CN105631872B (en) | Remote sensing image registration method based on multi-characteristic points | |
| CN106127258B (en) | A kind of target matching method | |
| CN108346162A (en) | Remote sensing image registration method based on structural information and space constraint | |
| CN103136760B (en) | A kind of multi-sensor image matching process based on FAST Yu DAISY | |
| CN108765476A (en) | A kind of polarization image method for registering | |
| Zhang et al. | Mutual information based multi-modal remote sensing image registration using adaptive feature weight | |
| CN107240130A (en) | Remote Sensing Image Matching method, apparatus and system | |
| CN107490356A (en) | A kind of noncooperative target rotary shaft and rotation angle measuring method | |
| CN114743189A (en) | Pointer instrument reading identification method and device, electronic equipment and storage medium | |
| CN106709870A (en) | Close-range image straight-line segment matching method | |
| CN104268550B (en) | Feature extracting method and device | |
| CN108257153A (en) | A kind of method for tracking target based on direction gradient statistical nature | |
| CN104820992B (en) | A kind of remote sensing images Semantic Similarity measure and device based on hypergraph model | |
| CN113628261B (en) | Infrared and visible light image registration method in electric power inspection scene |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |