CN101840517A - Control point image database matching method based on image registration and device thereof - Google Patents

Control point image database matching method based on image registration and device thereof Download PDF

Info

Publication number
CN101840517A
CN101840517A CN 201010161169 CN201010161169A CN101840517A CN 101840517 A CN101840517 A CN 101840517A CN 201010161169 CN201010161169 CN 201010161169 CN 201010161169 A CN201010161169 A CN 201010161169A CN 101840517 A CN101840517 A CN 101840517A
Authority
CN
China
Prior art keywords
image
point
reference mark
matching
unique point
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN 201010161169
Other languages
Chinese (zh)
Other versions
CN101840517B (en
Inventor
江万寿
岳春宇
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Wuhan University WHU
Original Assignee
Wuhan University WHU
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Wuhan University WHU filed Critical Wuhan University WHU
Priority to CN201010161169.7A priority Critical patent/CN101840517B/en
Publication of CN101840517A publication Critical patent/CN101840517A/en
Application granted granted Critical
Publication of CN101840517B publication Critical patent/CN101840517B/en
Expired - Fee Related legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Abstract

The invention relates to a control point image database matching method based on image registration, belonging to the field of surveying science and technique. First, the general range of regions to be matched in images to be matched is determined on the basis of control point image geographical coordinate information, and geometric distortion correction is conducted to the regions to be matched to obtain regions to be registered. Then, step matching strategies are adopted for matching, characteristic points are matched through a relaxation method, image registration is conducted to control point images and the regions to be registered, and the information of the whole control point images is used to constrain the matching of the control point images. In registration results, the matching results of the characteristic points closest to the centers of the control point images are used as final matching results. By adopting the method for matching the control points in the control point image database, good matching success rate can be obtained.

Description

A kind of reference mark image storehouse matching process and device thereof based on Image registration
Technical field
The invention belongs to the Surveying Science and Technology field, relate to a kind of image storehouse, reference mark matching process based on Image registration.This method is eliminated the image geometry distortion based on reference mark image geographic coordinate information; Adopt classification plan rate, the method for the lax coupling of unique point is carried out Image registration; Utilize the distance condition constraint to obtain final matching results at last.This method has the characteristics simple and practical, that matching probability is high.
Background technology
Geometric correction is a basic process of Flame Image Process, no matter is collinearity equation method or polynomial expression remedy, all needs to utilize ground control point to resolve transition matrix to realize geometric correction.Inefficiency is chosen at traditional reference mark, inorganizable and management.Therefore, it is just very necessary to realize choosing automatically of reference mark to set up image storehouse, reference mark.
Zhang Jixian etc. have early studied principle and the method for setting up large-scale reference mark database, have introduced how to extract the reference mark carry out relevant matches from database.People such as Chen Qihao have studied the inquiry mode at design, institutional framework and the reference mark in image storehouse, reference mark.Chen Gong etc. have proposed to utilize image storehouse, reference mark to carry out image to mate automatically, estimate the reliability of match point then according to related coefficient, and adopt some position prediction and pyramid strategy to improve search speed, considered the rotational deformation of image blocks, made registration reach subpixel accuracy by introducing the least square coupling.The geographical slip-stick artist of institute cloth Mr. Lan of France cooperates with SEP company, in the VIPS system, develop digitizing and obtained reference mark sub-image and the method for setting up the image storehouse, choose the reference mark image by map, and carry out the geographic coordinate measurement and obtain coordinate information, the coordinate range of the image that can correct according to need accesses control corresponding point sub-image by relevant automatically then, finds the position at reference mark on the new satellite image.
Though carried out many-sided research, still existed not enough in actual applications to choosing automatically based on the reference mark in image storehouse, reference mark.The image that the design in image storehouse, most of reference mark does not have to consider to have rotation and dimensional variation lacks the adaptability to different proportion chi and orientation image; Only depend on the image information at isolated point place based on the image matching strategy of single-point (part), be difficult for eliminating many couplings or false matching problem; The tradition correlation coefficient process can't be applied to inconsistent image of radiation characteristic or some position.
Summary of the invention
Problem to be solved by this invention is: a kind of image storehouse, reference mark matching process based on Image registration is provided.This method is at first determined the approximate range in the zone to be matched in the image to be matched based on reference mark image geographic coordinate information, and treats matching area and carry out the geometric distortion correction, obtains zone subject to registration.Adopt the classification matching strategy then, unique point relaxation method coupling is carried out Image registration to reference mark image and zone subject to registration, and the information of utilizing view picture reference mark image retrains the coupling of reference mark image.In registration results, the chosen distance reference mark immediate Feature Points Matching result of image center is as final matching results.Mate at the reference mark of using this method to carry out in the image storehouse, reference mark, has obtained and be matched to power preferably.
A kind of image storehouse, reference mark matching process based on Image registration is characterized in that, may further comprise the steps:
Step 1 is the geometric distortion of eliminating image to be matched with reference to image by the geometric distortion cancellation module with the reference mark image, obtains the zone subject to registration of pyramid image coupling;
Step 2, the lax coupling of classification is carried out in the zone subject to registration after by the pyramid image matching module reference mark image and step 1 being finished;
Step 3 concentrates image center nearest matching result in chosen distance reference mark as final matching results by the matching result point of matching result acquisition module after step 2 is finished.
At above-mentioned a kind of image storehouse, reference mark matching process based on Image registration, in the described step 1, concrete operation steps is as follows:
Step 1.1, parameter acquiring unit read image coordinate and resolve function and elevation datum: promptly parameter acquiring unit obtains reference mark image sensor imaging geometry model coordinate normal solution function f by image parameters C, image sensor imaging geometry model coordinate normal solution function f to be matched P, the anti-function f of separating of reference mark image sensor imaging geometry model coordinate C', the anti-function f of separating of image sensor imaging geometry model coordinate to be matched P', and image dispersed elevation value Z to be matched P, make Z PBe elevation datum.
Step 1.2 is read the image coordinate (x of four angle points of reference mark image by regional acquiring unit subject to registration Ci, y Ci), regional P in four angle points besieged city then CWith regional P CExpand k pixel, obtain a regional P C', be used for calculating the regional extent to be matched in image to be matched, P C' four angular coordinates are (x C1-k, y C1-k), (x C2+ k, y C2-k), (x C3-k, y C3+ k), (x C4+ k, y C4+ k);
Step 1.3, by the perspective transform parameter acquiring unit with the regional P in the step 1.2 C' four angular coordinate (x C1-k, y C1-k), (x C2+ k, y C2-k), (x C3-k, y C3+ k), (x C4+ k, y C4+ k) and step 1.1 in height value benchmark face amount Z PThe geometric model coordinate normal solution function f of reference mark image in the substitution step 1.1 CMiddle ground coordinate (the X that calculates four angular coordinate correspondences Pi, Y Pi), i.e. (X Pi, Y Pi)=f C(x Ci, y Ci, Z P), obtain the ground region P that is covered by reference mark image expansion back;
The anti-function of separating of geometric model coordinate with four angle point ground coordinate substitution images to be matched of the earth coverage area territory P that tries to achieve
Figure GSA00000110834200031
Calculate their corresponding image coordinates on image to be matched, (x Pi, y Pi)=f P' (X Pi, X Pi, Z R), the zone to be matched described in the above-mentioned steps 1.2 is (x Pi, y Pi) zone that surrounds, then with (x C1-k, y C1-k), (x C2+ k, y C2-k), (x C3-k, y C3+ k), (x C4+ k, y C4+ k) and (x Pi, y Pi) substitution perspective transform formula
x ′ = l 1 x + l 2 y + l 3 l 7 x + l 8 y + 1
y ′ = l 4 x + l 5 y + l 6 l 7 x + l 8 y + 1 , wherein l 1 l 2 l 3 l 4 l 5 l 6 l 7 l 8 1 ≠ 0
Promptly obtain l 1L 88 perspective transform parameters, i=1 wherein, 2,3,4;
Step 1.4, regional acquiring unit subject to registration carries out perspective transform with eight perspective transform parameters that obtain in the step 1.3 to overlay area, the ground respective regions of P on image to be matched, adopt the B-spline function method to resample to non-integer point, eliminate geometric distortion, obtain the imagery zone after the conversion, be the zone subject to registration of pyramid image coupling.
At above-mentioned a kind of image storehouse, reference mark matching process based on Image registration, in the described step 2, concrete steps are as follows:
Step 2.1 obtains the pyramid image matching parameter by the matching parameter acquiring unit, and the pyramid image matching parameter comprises scaling and pyramid progression, and concrete grammar is: pyramid image interlayer scaling P is set s=1/2 2n, n=1～3, first order pyramid image size S Min〉=64 * 64 pixels, first order pyramid image are the minimum image of top layer resolution in the pyramid structure; Sampling number
Figure GSA00000110834200036
Integral part, wherein L and H are the length and the height of reference mark image; Pyramid image coupling progression is N R+ 1;
Step 2.2 is carried out pyramid first order coupling by first order matching unit in the zone subject to registration that step 1 is finished;
Step 2.3, second level matching unit are carried out pyramid second level coupling in the zone subject to registration that step 1 is finished;
Step 2.4, N level matching unit are carried out pyramid N level coupling in the zone subject to registration that step 1 is finished, method is carried out the pyramid image coupling, until the afterbody pyramid image step by step with in the step 2.3;
Step 2.5, unique point and its matching result of record afterbody pyramid coupling.
At above-mentioned a kind of image storehouse, reference mark matching process based on Image registration, in the described step 2.2, concrete steps are as follows:
Step 2.21 is according to scaling
Figure GSA00000110834200037
To reference mark image and area resample subject to registration, generate reference mark image first order pyramid image and regional first order pyramid image subject to registration;
Step 2.22 is provided with match window W M1Size is length of side W A1The square window of=3 or 5 or 7 or 9 or 11 pixels, hunting zone S s=S LPixel * S HPixel, regional first order pyramid image subject to registration is long to be PL 1The pixel height is PH 1Pixel, reference mark image first order pyramid image is long to be OL 1The pixel height is OH 1Pixel, S L=PL 1-OL 1, S H=PH 1-OH 1
Step 2.23 is provided with parallactic grid and reliability graticule mesh on the image first order pyramid image of reference mark, parallactic grid and reliability graticule mesh constitute by square net, with coordinate (W A1/ 2+1, W A1/ 2+1) be starting point, the summit of square net is a grid points, the square net length of side is W A1, square net number=(OL 1/ W A1) * (OH 1/ W A1), be 0 with each graticule mesh point value assignment of parallactic grid;
Step 2.24 is used Shen Jun operator extract minutiae on the first order pyramid image of reference mark image; And calculate the gradient information on raw video of each unique point: comprise Grad
Figure GSA00000110834200041
Gradient direction angle gradDir=arctan ((g (x, y+1)-g (x, y))/(g (x+1, y)-g (x, y)))/π, wherein (x is that (x, the gray-scale value of y) locating judge whether gradient direction angle g to the image mid point y) to g De＞0.7 radian or g De＜0.3 radian is this unique point reliability value R then B=1, otherwise this unique point reliability value R B=0;
Step 2.25 is calculated the first candidate matches peak point;
Step 2.26, unique point adds the reliability graticule mesh, that is: according to the unique point coordinate, in each grid scope of reliability graticule mesh, reliability value R B＞0 unique point adds the net point array of this grid, and the net point array is recorded in characteristic point coordinates in this grid scope;
Step 2.27 is used existing all reliability value R B＞0 unique point makes up the Delaunay triangulation network;
Step 2.28 adopts the relaxation method feature matching method to all reliability value R B＞0 the unique point coupling that relaxes, the concrete grammar of lax coupling is: optimize the first candidate matches peak point of each unique point again, and (dx dy) makes it equal the parallax parameter (md of the first candidate matches peak point to optimize the parallax parameter of unique point x, md y); If the facies relationship numerical value of the first candidate matches peak point of unique point makes this unique point reliability value R greater than 0.75 B=2;
Step 2.29, multinomial model is optimized the first candidate matches peak point of unique point and is preserved parallactic grid at the corresponding levels.
At above-mentioned a kind of image storehouse, reference mark matching process based on Image registration, in the described step 2.25, the concrete steps of calculated candidate match peak point are as follows:
Step 2.251, the screening unique point: the unique point to extracting judges whether reliability value R B＞0, be then to carry out process steps 2.253, otherwise carry out process steps 2.252;
Step 2.252 is rejected this unique point, chooses next unique point, changes step 2.51;
Step 2.253 is calculated related coefficient, and concrete steps are as follows:
Step a, (x y) is the center, according to match window size W with the unique point coordinate on the image first order pyramid image of reference mark M1, calculate the target area, the target area is and match window square of the same size that its upper left angle point is (x-W A1/ 2, y-W A1/ 2) the bottom right angle point is (x+W A1/ 2, y+W A1/ 2);
Step b, (x y) is the center, according to described search range S with the unique point coordinate on regional first order pyramid image subject to registration s, calculate the region of search, the region of search is a rectangle, its upper left angle point is (x-S L/ 2, y-S H/ 2), the bottom right angle point is (x+S L/ 2, y+S H/ 2);
Step c in the region of search, overlaps the position with the upper left angle point in target area and region of search and begins, moving object region, calculating related coefficient ρ (c, r):
ρ ( c , r ) = Σ i = x - w a 1 / 2 w a 1 Σ j = y - w a 1 / 2 w a 1 ( g i , j - g ‾ ) ( g i + r , j + c ′ - g ‾ ′ ) Σ i = x - w a 1 / 2 w a Σ j = y - w a 1 / 2 w a ( g i , j - g ‾ ) 2 × ( g i + r , j + c ′ - g ‾ r , c ′ ) 2
Whenever move and once be a match window, write down the facies relationship numerical value of each match window;
Wherein c, the row, column displacement when r is moving object region on the region of search, g I, jFor in the target area (i, gray-scale value j),
Figure GSA00000110834200052
Be the target area average gray value,
Figure GSA00000110834200053
For in the region of search (i+r, gray-scale value j+c),
Figure GSA00000110834200054
For moved on the region of search when the target area displacement (c, in the region of search of participate in calculating in the time of r) with the average gray value in the big or small corresponding zone of match window;
Step 2.254 is calculated parallax parameter;
Matching result parallax parameter (the md of each match window x, md y) be:
md x＝x c-x，
md y＝y c-y，
Wherein, (x c, y c) be this match window center point coordinate, (x y) is current unique point coordinate;
Related coefficient and parallax parameter are charged to matching result array BP Cnds, the matching result of each match window is described by facies relationship numerical value and parallax parameter;
Step 2.255 is determined the first candidate matches peak point of unique point;
In the region of search, the facies relationship numerical value that compares the match window of each match window and its eight neighborhood, judge that whether this match window related coefficient is all greater than its eight neighborhood matching window-related coefficient, be then with this match window central point as the candidate matches peak point, carry out process steps 2.56,, otherwise current unique point reliability value R BBe made as 0, turn over journey step 2.52;
Step 2.256, all candidate matches peak points according to related coefficient size ordering, the maximum first candidate matches peak point of classifying as;
Step 2.257 is chosen next unique point, changes as step 2.251, if all unique points are all calculated and finished, enters above-mentioned steps 2.26.
At above-mentioned a kind of image storehouse, reference mark matching process based on Image registration, in the described step 2.29, the concrete steps of the first candidate matches peak point of multinomial model optimization unique point are as follows:
Step 2.291 is calculated the polynomial fitting coefficient, and the affined transformation model that uses in the calculating is as follows:
md x＝a 1+a 2×xd+a 3×yd，
md y＝b 1+b 2×xd+b 3×yd，
Wherein, md x, md yBe the parallax parameter of the first candidate matches peak point of unique point, xd=x-xc, yd=y-yc, x, y are characteristic point coordinates, xc, yc are the parallactic grid center point coordinate; With all R in the feature point set B＞1 unique point substitution affined transformation model, each feature point range one set of equations is calculated a by least square method 1, a 2, a 3, b 1, b 2, b 3Six affined transformation coefficients;
Step 2.292, optimize the first candidate matches peak point:
Steps A, calculated characteristics is put the deviate of the first candidate matches peak point, and concrete grammar is as follows:
Make updating mark f=0; Judge whether current unique point R in the feature point set B＞1, otherwise change step F, be then unique point substitution polynomial fitting to be calculated:
px＝a 1+a 2×xd+a 3×yd，
py＝b 1+b 2×xd+b 3×yd，
Wherein, px, py parallax value in polynomial fitting, calculating, xd=x-xc, yd=y-yc, x, y are characteristic point coordinates, xc, yc are the parallactic grid center point coordinate;
Calculation deviation value t again x, t y:
t x=| px-md x|, t y=| py-md y|; Wherein, md x, md yParallax parameter for the first candidate matches peak point of unique point;
Step B judges deviate: judge t x＞12 and t y＞12, be then to carry out step C, then change step F if not over to;
Step C, calculate the deviation of next candidate matches peak point: judging whether next candidate matches peak point of this unique point exists, is then to change step D; Otherwise commentaries on classics step e;
Step D upgrades the first candidate matches peak point; Concrete grammar is: make updating mark f=1; According to the parallax parameter of this candidate matches peak point, calculate new deviate t Xi, t Yi
Judge whether t Xi＜t xAnd t Yi＜t y, otherwise rotor step C is then to make t x=t Xi, t y=t Yi, as the first candidate matches peak point, its parallax parameter value is composed parallax parameter to unique point, and (dx dy), and makes the reliability value R of this unique point with this candidate matches peak point B=1, change step C;
Step e judges whether updating mark f=1, is then to change step F over to; Otherwise make this unique point reliability value R B=0, change step F;
Step F, judging characteristic point concentrates whether there is next unique point, and being then changes steps A, otherwise changes step 2.293;
Step 2.293 replacement parallactic grid grid value: each graticule mesh point value of parallactic grid (mx my) is:
Mx=a 1+ a 2* Sx+a 3* Sy, my=b 1+ b 2* Sx+b 3* Sy, wherein, Sx, Sy are the distance of this grid points to the graticule mesh center.
At above-mentioned a kind of image storehouse, reference mark matching process based on Image registration, in the described step 3, final matching results determines that method is as follows:
If the unique point of final entry in the step 2.5 is reliability value R B＞1, then calculate its image coordinate (x, y) to the Euclidean distance m of reference mark image center (x ', y '),
Figure GSA00000110834200071
Travel through all unique points, obtain minimum m value, promptly nearest apart from reference mark image center unique point is as the reference mark matching result.
A kind of device that adopts based on image storehouse, the reference mark matching process of Image registration, comprise geometric distortion cancellation module and the pyramid image matching module that links to each other with the geometric distortion cancellation module, the matching result acquisition module links to each other with described pyramid image matching module.
At above-mentioned a kind of device that adopts based on image storehouse, the reference mark matching process of Image registration, described geometric distortion cancellation module comprises parameter acquiring unit, regional acquiring unit subject to registration and the perspective transform parameter acquiring unit that links to each other successively.
At above-mentioned a kind of device that adopts based on image storehouse, the reference mark matching process of Image registration, described pyramid image matching module comprises matching parameter acquiring unit, first order matching unit, second level matching unit and the N level matching unit that links to each other successively.
The technique effect of this Fang Ming is embodied in: at first determine to be matched regional approximate range in the image to be matched based on reference mark image geographic coordinate information, and treat matching area and carry out geometric distortion and correct, obtain zone subject to registration.Adopt the classification matching strategy then, unique point relaxation method coupling is carried out Image registration to reference mark image and zone subject to registration, and the information of utilizing view picture reference mark image retrains the coupling of reference mark image.In registration results, the chosen distance reference mark immediate Feature Points Matching result of image center is as final matching results.Mate at the reference mark of using this method to carry out in the image storehouse, reference mark, has obtained and be matched to power preferably.
Description of drawings
Fig. 1 is an operational flowchart of the present invention;
Fig. 2 is image geometry distortion removing method synoptic diagram of the present invention;
Fig. 3 is a parallactic grid synoptic diagram of the present invention;
Fig. 4 is a reliability graticule mesh synoptic diagram of the present invention;
Fig. 5 is an apparatus structure synoptic diagram of the present invention.
Embodiment
Below in conjunction with accompanying drawing the present invention is done to describe in further detail, this method flow process is seen Fig. 1, and summary is got up, and the enforcement of this method can be divided into three steps:
1. eliminate the image geometry distortion
Resolve function and elevation datum 1.1 read image coordinate
Read the image parameters of reference mark image and image to be matched.The reference mark image has Geographical Coordinates Parameter among the embodiment, and image to be matched has the RPC parameter.
Obtain reference mark image sensor imaging geometry model coordinate normal solution function f by image parameters C, image sensor imaging geometry model coordinate normal solution function f to be matched P, the anti-function f of separating of reference mark image sensor imaging geometry model coordinate C', the anti-function f of separating of image sensor imaging geometry model coordinate to be matched P', and image dispersed elevation value Z to be matched P, make Z PBe elevation datum.
1.2 calculate image resampling parameter to be matched
Read the image coordinate (x of four angle points of reference mark image Ci, y Ci), because the reference mark image is a regular rectangular shape, rectangular area P in four angle points besieged city then C
This rectangular area is extended out 100 pixels, obtain a bigger rectangular area P C', be used for calculating the zone to be matched in image to be matched.Four angular coordinates of this rectangular area are (x C1-100, y C1-100), (x C2+ 100, y C2-100), (x C3-100, y C3+ 100), (x C4+ 100, y C4+ 100).
These four coordinates and height value benchmark face amount Z PThe geometric model coordinate normal solution function f of substitution reference mark image CCalculate the ground coordinate (X of these four some correspondences Pi, Y Pi), i.e. (X Pi, Y Pi)=f C(x Ci, y Ci, Z P).Obtain expanding the ground region P that the back is covered by the reference mark image.
The anti-function of separating of geometric model coordinate of four angle point ground coordinate substitution images to be matched of the earth coverage area territory P that tries to achieve
Figure GSA00000110834200091
Calculate their corresponding image coordinate, i.e. (x on right image Pi, y Pi)=f P' (X Pi, X Pi, Z R).I=1 wherein, 2,3,4.
(x C1-100, y C1-100), (x C2+ 100, y C2-100), (x C3-100, y C3+ 100), (x C4+ 100, y C4+ 100) and (x Pi, y Pi) substitution perspective transform formula
x ′ = l 1 x + l 2 y + l 3 l 7 x + l 8 y + 1
y ′ = l 4 x + l 5 y + l 6 l 7 x + l 8 y + 1 , wherein l 1 l 2 l 3 l 4 l 5 l 6 l 7 l 8 1 ≠ 0
But 2 equations of every group of coordinate row, but four groups of coordinates are total to 8 equations of row, and computational solution gets l 1L 88 perspective transform parameters.
1.3 overlay area, the ground respective regions of P on image to be matched carried out perspective transform by eight perspective transform parameters that 1.2 steps calculated, adopt the B-spline function method to resample to non-integer point, eliminate geometric distortion, obtain the imagery zone after the conversion, the zone subject to registration when being the pyramid coupling.
Eliminate geometric distortion, obtain zone subject to registration method see Fig. 2.
2 reference mark Image registrations
Gold tower matching strategy is adopted in the reference mark Image registration
2.1 calculate the pyramid image matching parameter
Image size in reference mark is 512 pixels * 512 pixels among the embodiment, and the region of search size is (512 pixels+200 pixels) * (512 pixels+200 pixels) in the image to be matched
Be provided with that scaling is 1/4 between pyramidal layer, sampling number
Figure GSA00000110834200096
Pyramid image coupling progression is N R+ 1=4;
2.2 pyramid classification coupling
2.2.1 first order coupling
(1) according to scaling 1/4 3,, generate reference mark image first order pyramid image and first order pyramid image subject to registration to reference mark image and area resample subject to registration;
(2) the match window size being set is 5 pixels * 5 pixels, and the hunting zone is long to be (512 pixels+200 pixels)/2 3-512 pixels/2 3=25 pixels, height are (512 pixels+200 pixels)/2 3-512 pixels/2 3=25 pixels;
(3) parallactic grid and reliability graticule mesh are set;
On the image first order pyramid image of reference mark, parallactic grid and reliability graticule mesh are set, all with (3,3) be starting point, be spaced apart match window size 5 pixels * 5 pixels, meshes number=(64 pixels/5 pixels) * (64 pixels/5 pixels)=12 * 12, see Fig. 3 and Fig. 4, wherein circle is represented grid points among Fig. 3;
The graticule mesh point value initialize of parallactic grid is 0;
(4) extract minutiae
(4.1) use Shen Jun operator extract minutiae collection on the image first order pyramid image of reference mark;
(4.2) calculated characteristics point gradient direction angle g DeWith Grad g d
g d = ( g ( x + 1 , y ) - g ( x , y ) ) 2 + ( g ( x , y + 1 ) - g ( x , y ) ) 2 ,
g de＝arctan((g(x，y+1)-g(x，y))/(g(x+1，y)-g(x，y)))/π，
Wherein (x y) is unique point (x, gray-scale value y) in the image first order pyramid image of reference mark to g;
(4.3) judge whether gradient direction angle g De＞0.7 radian or g De＜0.3 radian is this unique point reliability value R then B=1, otherwise this unique point reliability value R B=0;
(5) calculate the first candidate matches peak point,
(5.1) screening unique point;
Unique point to extracting judges whether reliability value R B＞0, be the process of then carrying out (5.3), otherwise the process of carrying out (5.2);
(5.2) reject this unique point, in feature point set, choose next unique point, turn over journey (5.1); Should be noted in the discussion above that the present invention only is applicable to that remaining the next one at least after this step is rejected meets R BThe situation of＞0 unique point, if all unique points are all disallowable, then this method lost efficacy;
(5.3) calculate related coefficient;
(x y) is the center, calculates the target area with the unique point coordinate on the image first order pyramid image of reference mark, the target area is and the consistent square of match window size (5 pixels * 5 pixels), its upper left angle point be (x-2, y-2) the bottom right angle point be (x+2, y+2);
(x y) is the center, calculates the region of search with the unique point coordinate on regional first order pyramid image subject to registration, the region of search is and the consistent rectangle of search range (25 pixels * 25 pixels), its upper left angle point be (x-12, y-12) the bottom right angle point be (x+12, y+12);
In the region of search, overlap the position with the upper left angle point in target area and region of search and begin, moving object region, calculating related coefficient ρ (c, r):
ρ ( c , r ) = Σ i = x - w a 1 / 2 w a 1 Σ j = y - w a 1 / 2 w a 1 ( g i , j - g ‾ ) ( g i + r , j + c ′ - g ‾ ′ ) Σ i = x - w a 1 / 2 w a Σ j = y - w a 1 / 2 w a ( g i , j - g ‾ ) 2 × ( g i + r , j + c ′ - g ‾ r , c ′ ) 2
Whenever move and once be a match window, write down the facies relationship numerical value of each match window;
Wherein c, the row, column displacement when r is moving object region on the region of search, g I, jFor in the target area (i, gray-scale value j),
Figure GSA00000110834200112
Be the target area average gray value,
Figure GSA00000110834200113
For in the region of search (i+r, gray-scale value j+c), For moved on the region of search when the target area displacement (c, in the region of search of participate in calculating in the time of r) with the average gray value in the big or small corresponding zone of match window;
(5.4) calculate parallax parameter;
Matching result parallax parameter (the md of each match window x, md y) be:
md x＝x c-x，md y＝y c-y，
Wherein, x, y are current unique point coordinate, x c, y cBe this match window center point coordinate;
The matching result of each match window is described by facies relationship numerical value and parallax parameter, and the related coefficient and the parallax parameter of each match window are charged to matching result array BP Cnds
(5.5) determine the first candidate matches peak point of unique point;
In the region of search, the facies relationship numerical value that compares the match window of each match window and its eight neighborhood, whether judge this match window related coefficient all greater than its eight neighborhood matching window-related coefficient, be then with this match window central point as the candidate matches peak point;
Traveling through whole match windows, judge whether to exist the candidate matches peak point, is the process of then carrying out (5.6), otherwise current unique point reliability value R BBe made as 0, turn over journey (5.2);
(5.6) with all candidate matches peak points according to related coefficient size ordering, the maximum first candidate matches peak point of classifying as;
(5.7) choose next unique point, turn over journey (5.1),, enter substep (6) if all unique points are all calculated and finished;
(6) unique point adds the reliability graticule mesh;
According to the unique point coordinate, in each grid scope of reliability graticule mesh, reliability value R B＞0 unique point adds the net point array of this grid, and the net point array is recorded in characteristic point coordinates in this grid scope;
(7) use existing all reliability value R B＞0 unique point makes up the Delaunay triangulation network;
(8) by the relaxation method feature matching method of introducing in the document " research of aviation image characteristic matching " (Jiang Wanshou etc., Wuhan University's journal information science version, 28 (5), 2003) to all reliability value R B＞0 the unique point coupling that relaxes,
The corresponding candidate matches peak value point set of the present invention of candidate matches point set wherein;
Again optimize the first candidate matches peak point of each unique point, and (dx dy) makes it equal the parallax parameter (md of the first candidate matches peak point to optimize the parallax parameter of unique point x, md y); If the facies relationship numerical value of the first candidate matches peak point of unique point makes this unique point reliability value R greater than 0.75 B=2.
(9) multinomial model is optimized its first candidate matches peak point of unique point
(9.1) calculate the polynomial fitting coefficient
Use the affined transformation model:
md x＝a 1+a 2×xd+a 3×yd，
md y＝b 1+b 2×xd+b 3×yd，
Wherein, md x, md yBe the parallax parameter of the first candidate matches peak point of unique point, xd=x-xc, yd=y-yc, x, y are characteristic point coordinates, xc, yc are the parallactic grid center point coordinate,
With all R in the feature point set B＞1 unique point substitution affined transformation model, each feature point range one set of equations is calculated a by least square method 1, a 2, a 3, b 1, b 2, b 3Six affined transformation coefficients;
(9.2) optimize the first candidate matches peak point
(9.2.1) calculated characteristics is put the deviate of the first candidate matches peak point
Make updating mark f=0;
Judge whether current unique point R in the feature point set B＞1, otherwise rotor process (9.2.6) is then unique point substitution polynomial fitting to be calculated:
px＝a 1+a 2×xd+a 3×yd，
py＝b 1+b 2×xd+b 3×yd，
Wherein, px, py parallax value in polynomial fitting, calculating, xd=x-xc, yd=y-yc, x, y are characteristic point coordinates, xc, yc are the parallactic grid center point coordinate;
Calculation deviation value t again x, t y:
t x＝|px-md x|，t y＝|py-md y|；
Wherein, md x, md yParallax parameter for the first candidate matches peak point of unique point;
(9.2.2) judge deviate
Judge t x＞12 and t y＞12, be the process of then carrying out (9.2.3), then change process (9.2.6) if not over to;
(9.2.3) calculate the deviation of next candidate matches peak point
Judging whether next candidate matches peak point of this unique point exists, is rotor process (9.2.4) then; Otherwise rotor process (9.2.5);
(9.2.4) upgrade the first candidate matches peak point;
Make updating mark f=1;
According to the parallax parameter of this candidate matches peak point, calculate new deviate t Xi, t Yi
Judge whether t Xi＜t xAnd t Yi＜t y, otherwise rotor process (9.2.3) is then to make t x=t Xi, t y=t Yi, as the first candidate matches peak point, its parallax parameter value is composed parallax parameter to unique point, and (dx dy), and makes the reliability value R of this unique point with this candidate matches peak point B=1, rotor process (9.2.3);
(9.2.5) judge whether updating mark f=1, be rotor process (9.2.6) then, otherwise make this unique point reliability value R B=0, rotor process (9.2.6);
(9.2.6) judging characteristic point is concentrated and whether to be had next unique point, is rotor process (9.2.1) then, otherwise turns over journey (9.3);
(9.3) replacement parallactic grid grid value
Each graticule mesh point value of parallactic grid (mx my) is:
mx＝a 1+a 2×Sx+a 3×Sy，
my＝b 1+b 2×Sx+b 3×Sy，
Wherein, Sx, Sy are the distance of this grid points to the graticule mesh center;
(10) preserve parallactic grid at the corresponding levels;
2.2.2 second level coupling
(1) according to scaling
Figure GSA00000110834200131
To reference mark image and area resample subject to registration, generate reference mark image second level pyramid image and the regional second level subject to registration pyramid image, the image second level, reference mark pyramid image is long to be OL 2The pixel height is OH 2Pixel, in the present embodiment, according to zoom ratio 1/4 2To reference mark image and area resample subject to registration, generate reference mark image second level pyramid image and the regional second level subject to registration pyramid image, the image second level, reference mark pyramid image size is (512 pixels/4) * (512 pixels/4)=128 pixels * 128 pixels, and pyramid image size in the regional second level subject to registration is (712 pixels/4) * (712 pixels/4)=178 pixels * 178 pixels;
(2) match window W is set M2Size is length of side W A2The square window of=3,5 or 7 pixels, hunting zone are length of side S A2=7～15 square area; In the present embodiment, it is 5 pixels * 5 pixels that the match window size is set, and the hunting zone is 11 pixels * 11 pixels;
(3) second level parallactic grid and second level reliability graticule mesh are set on the pyramid image of the image second level, reference mark, second level parallactic grid is identical with second level reliability grid structure, constitutes by square net, all with coordinate (W A2/ 2+1, W A2/ 2+1) be starting point, the summit of square net is a grid points, the square net length of side is W A2, square net number=(OL 2/ W A2) * (OH 2/ W A2); In the present embodiment, second level parallactic grid and second level reliability graticule mesh are set on the pyramid image of the image second level, reference mark, are starting point with coordinate (3,3) all, interval 5 pixels * 5 pixels, meshes number=(128 pixels/5 pixels) * (128 pixels/5 pixels)=25 * 25;
The grid points coordinate of second level parallactic grid multiply by scaling 1/4, and substitution first order parallactic grid is calculated the graticule mesh point value of second level parallactic grid by each graticule mesh point value bilinear interpolation of first order parallactic grid;
(4), and unique point dependability parameter R is set according to substep among the 2.2.1 (4) extract minutiae collection on the pyramid image of the image second level, reference mark BValue;
(5) with in the parallactic grid of the unique point coordinate substitution second level, bilinear interpolation is calculated the parallax value on its x direction and the y direction in the graticule mesh that unique point falls into, as the parallax parameter of this unique point (dx, dy);
(6) determine the region of search of each Feature Points Matching;
On the pyramid image of the image subject to registration second level with unique point coordinate (x, y) be the center, calculate the region of search, the region of search is and the consistent square of the described search range of substep (2), with characteristic point coordinates (x, y) deduct unique point parallax parameter (dx, dy) after, be the center, calculate the region of search:
x from＝x-dx-S a2/2，y from＝y-dx-S a2/2；
x to＝x-dx+S a2/2，y to＝y-dy+S a2/2；
In the present embodiment, calculating the region of search is:
x from＝x-dx-5，y from＝y-dx-5；
x to＝x-dx+5，y to＝y-dy+5；
(x wherein From, y From) be the upper left angle point in region of search, (x To, y To) be bottom right, region of search angle point;
(7) according to substep among the 2.2.1 (5), calculated characteristics point is concentrated the candidate matches peak point of each unique point, and wherein match window is revised as W M2, it is S that the length of side is revised as in the hunting zone A2Square area; In the present embodiment, match window is revised as 5 pixels * 5 pixels, and 11 pixels * 11 pixels are revised as in the hunting zone;
(8) carry out the lax coupling of second level pyramid according to 2.2.1 (6)-(10) step, and preserve second level parallactic grid;
2.2.3 third level coupling
In the present embodiment, the concrete steps of third level coupling are as follows:
(1) according to zoom ratio 1/4, to reference mark image and area resample subject to registration, generate reference mark image third level pyramid image and regional third level pyramid image subject to registration, reference mark image third level pyramid image size is (512 pixels/2) * (512 pixels/2)=256 pixels * 256 pixels, and the regional third level subject to registration and word tower image size are (712 pixels/2) * (712 pixels/2)=356 pixels * 356 pixels;
(2) the match window size being set is 5 pixels * 5 pixels, and the hunting zone is 11 pixels * 11 pixels
(3) on the image third level pyramid image of reference mark, parallactic grid and reliability graticule mesh are set according to (3) among 2.2.2 step, and calculate each graticule mesh point value in the parallactic grid;
(4) carry out the lax coupling of third level pyramid according to (4) among the 2.2.2-(8) step, and preserve parallactic grid at the corresponding levels;
2.4 fourth stage coupling, in the present embodiment, the concrete steps of fourth stage coupling are as follows:
(1) reference mark image fourth stage pyramid image size is 512 * 512, and image fourth stage pyramid image size subject to registration is 712 * 712;
(2) the match window size being set is 5 pixels * 5 pixels, and the hunting zone is 11 pixels * 11 pixels;
(3) be located on the image fourth stage pyramid image of reference mark parallactic grid and reliability graticule mesh are set according to (3) among 2.2.2 step, and calculate each graticule mesh point value in the parallactic grid;
(4) carry out the lax coupling of third level pyramid according to (4) among the 2.2.1-(8) step, and preserve the unique point and the matching result thereof of pyramid coupling at the corresponding levels;
3. optimized choice matching result
If the unique point of final entry is reliability value R B＞1, then calculate its image coordinate (x, y) to the Euclidean distance m of reference mark image center (x ', y '), wherein,
Figure GSA00000110834200151
Travel through all unique points, obtain minimum m value, promptly nearest apart from reference mark image center unique point is as the reference mark matching result.
In the present embodiment, by the result such as the following table of the present invention's coupling
The reference mark sum Correct coupling Erroneous matching Be matched to power
40 32 8 80％

Claims (10)

1. image storehouse, the reference mark matching process based on Image registration is characterized in that, may further comprise the steps:
Step 1 is the geometric distortion of eliminating image to be matched with reference to image by the geometric distortion cancellation module with the reference mark image, obtains the zone subject to registration of pyramid image coupling;
Step 2, the lax coupling of classification is carried out in the zone subject to registration after by the pyramid image matching module reference mark image and step 1 being finished;
Step 3 concentrates image center nearest matching result in chosen distance reference mark as final matching results by the matching result point of matching result acquisition module after step 2 is finished.
2. a kind of image storehouse, reference mark matching process based on Image registration according to claim 1 is characterized in that in the described step 1, concrete operation steps is as follows:
Step 1.1, parameter acquiring unit read image coordinate and resolve function and elevation datum: promptly parameter acquiring unit obtains reference mark image sensor imaging geometry model coordinate normal solution function f by image parameters C, image sensor imaging geometry model coordinate normal solution function f to be matched P, the anti-function f of separating of reference mark image sensor imaging geometry model coordinate C', the anti-function f of separating of image sensor imaging geometry model coordinate to be matched P', and image dispersed elevation value Z to be matched P, make Z PBe elevation datum.
Step 1.2 is read the image coordinate (x of four angle points of reference mark image by regional acquiring unit subject to registration Ci, y Ci), regional P in four angle points besieged city then CWith regional P CExpand k pixel, obtain a regional P C', be used for calculating the regional extent to be matched in image to be matched, P C' four angular coordinates are (x C1-k, y C1-k), (x C2+ k, y C2-k), (x C3-k, y C3+ k), (x C4+ k, y C4+ k);
Step 1.3, by the perspective transform parameter acquiring unit with the regional P in the step 1.2 C' four angular coordinate (x C1-k, y C1-k), (x C2+ k, y C2-k), (x C3-k, y C3+ k), (x C4+ k, y C4+ k) and step 1.1 in height value benchmark face amount Z PThe geometric model coordinate normal solution function f of reference mark image in the substitution step 1.1 CMiddle ground coordinate (the X that calculates four angular coordinate correspondences Pi, Y Pi), i.e. (X Pi, Y Pi)=f C(x Ci, y Ci, Z P), obtain the ground region P that is covered by reference mark image expansion back;
The anti-function of separating of geometric model coordinate with four angle point ground coordinate substitution images to be matched of the earth coverage area territory P that tries to achieve
Figure FSA00000110834100021
Calculate their corresponding image coordinates on image to be matched,
Figure FSA00000110834100022
Zone to be matched described in the above-mentioned steps 1.2 is (x Pi, y Pi) zone that surrounds, then with (x C1-k, y C1-k), (x C2+ k, y C2-k), (x C3-k, y C3+ k), (x C4+ k, y C4+ k) and (x Pi, y Pi) substitution perspective transform formula
x ′ = l 1 x + l 2 y + l 3 l 7 x + l 8 y + 1 y ′ = l 4 x + l 5 y + l 6 l 7 x + l 8 y + 1 , Wherein l 1 l 2 l 3 l 4 l 5 l 6 l 7 l 8 1 ≠ 0
Promptly obtain l 1L 88 perspective transform parameters, i=1 wherein, 2,3,4;
Step 1.4, regional acquiring unit subject to registration carries out perspective transform with eight perspective transform parameters that obtain in the step 1.3 to overlay area, the ground respective regions of P on image to be matched, adopt the B-spline function method to resample to non-integer point, eliminate geometric distortion, obtain the imagery zone after the conversion, be the zone subject to registration of pyramid image coupling.
3. a kind of image storehouse, reference mark matching process based on Image registration according to claim 1 is characterized in that in the described step 2, concrete steps are as follows:
Step 2.1 obtains the pyramid image matching parameter by the matching parameter acquiring unit, and the pyramid image matching parameter comprises scaling and pyramid progression, and concrete grammar is: pyramid image interlayer scaling P is set s=1/2 2n, n=1～3, first order pyramid image size S Min〉=64 * 64 pixels, first order pyramid image are in the pyramid structure minimum image of layer resolution; Sampling number
Figure FSA00000110834100025
Integral part, wherein L and H are the length and the height of reference mark image; Pyramid image coupling progression is N R+ 1;
Step 2.2 is carried out pyramid first order coupling by first order matching unit in the zone subject to registration that step 1 is finished;
Step 2.3, second level matching unit are carried out pyramid second level coupling in the zone subject to registration that step 1 is finished;
Step 2.4, N level matching unit are carried out pyramid N level coupling in the zone subject to registration that step 1 is finished, method is carried out the pyramid image coupling, until the afterbody pyramid image step by step with in the step 2.3;
Step 2.5, unique point and its matching result of record afterbody pyramid coupling.
4. a kind of image storehouse, reference mark matching process based on Image registration according to claim 1 is characterized in that in the described step 2.2, concrete steps are as follows:
Step 2.21 is according to scaling
Figure FSA00000110834100031
To reference mark image and area resample subject to registration, generate reference mark image first order pyramid image and regional first order pyramid image subject to registration;
Step 2.22 is provided with match window W M1Size is length of side W A1The square window of=3 or 5 or 7 or 9 or 11 pixels, hunting zone S s=S LPixel * S HPixel, regional first order pyramid image subject to registration is long to be PL 1The pixel height is PH 1Pixel, reference mark image first order pyramid image is long to be OL 1The pixel height is OH 1Pixel, S L=PL 1-OL 1, S H=PH 1-OH 1
Step 2.23 is provided with parallactic grid and reliability graticule mesh on the image first order pyramid image of reference mark, parallactic grid and reliability graticule mesh constitute by square net, with coordinate (W A1/ 2+1, W A1/ 2+1) be starting point, the summit of square net is a grid points, the square net length of side is W A1, square net number=(OL 1/ Wa 1) * (OH 1/ W A1), be 0 with each graticule mesh point value assignment of parallactic grid;
Step 2.24 is used Shen Jun operator extract minutiae on the first order pyramid image of reference mark image; And calculate the gradient information on raw video of each unique point: comprise Grad Gradient direction angle gradDir=arctan ((g (x, y+1)-g (x, y))/(g (x+1, y)-g (x, y)))/π, wherein (x is that (x, the gray-scale value of y) locating judge whether gradient direction angle g to the image mid point y) to g De＞0.7 radian or g De＜0.3 radian is this unique point reliability value R then B=1, otherwise this unique point reliability value R B=0;
Step 2.25 is calculated the first candidate matches peak point;
Step 2.26, unique point adds the reliability graticule mesh, that is: according to the unique point coordinate, in each grid scope of reliability graticule mesh, reliability value R B＞0 unique point adds the net point array of this grid, and the net point array is recorded in characteristic point coordinates in this grid scope;
Step 2.27 is used existing all reliability value R B＞0 unique point makes up the Delaunay triangulation network;
Step 2.28 adopts the relaxation method feature matching method to all reliability value R B＞0 the unique point coupling that relaxes, the concrete grammar of lax coupling is: optimize the first candidate matches peak point of each unique point again, and (dx dy) makes it equal the parallax parameter (md of the first candidate matches peak point to optimize the parallax parameter of unique point x, md y); If the facies relationship numerical value of the first candidate matches peak point of unique point makes this unique point reliability value R greater than 0.75 B=2;
Step 2.29, multinomial model is optimized the first candidate matches peak point of unique point and is preserved parallactic grid at the corresponding levels.
5. a kind of image storehouse, reference mark matching process based on Image registration according to claim 2 is characterized in that in the described step 2.25, the concrete steps of calculated candidate match peak point are as follows:
Step 2.251, the screening unique point: the unique point to extracting judges whether reliability value R B＞0, be then to carry out process steps 2.253, otherwise carry out process steps 2.252;
Step 2.252 is rejected this unique point, chooses next unique point, changes step 2.51;
Step 2.253 is calculated related coefficient, and concrete steps are as follows:
Step a, (x y) is the center, according to match window size W with the unique point coordinate on the image first order pyramid image of reference mark M1, calculate the target area, the target area is and match window square of the same size that its upper left angle point is (x-W A1/ 2, y-W A1/ 2) the bottom right angle point is (x+W A1/ 2, y+W A1/ 2);
Step b, (x y) is the center, according to described search range S with the unique point coordinate on regional first order pyramid image subject to registration s, calculate the region of search, the region of search is a rectangle, its upper left angle point is (x-S L/ 2, y-S H/ 2), the bottom right angle point is (x+S L/ 2, y+S H/ 2);
Step c in the region of search, overlaps the position with the upper left angle point in target area and region of search and begins, moving object region, calculating related coefficient ρ (c, r):
ρ ( c , r ) = Σ i = x - w a 1 / 2 w a 1 Σ j = y - w a 1 / 2 w a 1 ( g i , j - g ‾ ) ( g i + r , j + c ′ - g ‾ ′ ) Σ i = x - w a 1 / 2 w a Σ j = y - w a 1 / 2 w a ( g i , j - g ‾ ) 2 × ( g i + r , j + c ′ - g r , c ‾ ′ ) 2
Whenever move and once be a match window, write down the facies relationship numerical value of each match window;
Wherein c, the row, column displacement when r is moving object region on the region of search, g I, jFor in the target area (i, gray-scale value j),
Figure FSA00000110834100052
Be the target area average gray value,
Figure FSA00000110834100053
For in the region of search (i+r, gray-scale value j+c),
Figure FSA00000110834100054
For moved on the region of search when the target area displacement (c, in the region of search of participate in calculating in the time of r) with the average gray value in the big or small corresponding zone of match window;
Step 2.254 is calculated parallax parameter;
Matching result parallax parameter (the md of each match window x, md y) be:
md x＝x c-x，
md y＝y c-y，
Wherein, (x c, y c) be this match window center point coordinate, (x y) is current unique point coordinate; Related coefficient and parallax parameter are charged to matching result array BP Cnds, the matching result of each match window is described by facies relationship numerical value and parallax parameter;
Step 2.255 is determined the first candidate matches peak point of unique point;
In the region of search, the facies relationship numerical value that compares the match window of each match window and its eight neighborhood, judge that whether this match window related coefficient is all greater than its eight neighborhood matching window-related coefficient, be then with this match window central point as the candidate matches peak point, carry out process steps 2.56,, otherwise current unique point reliability value R BBe made as 0, turn over journey step 2.52;
Step 2.256, all candidate matches peak points according to related coefficient size ordering, the maximum first candidate matches peak point of classifying as;
Step 2.257 is chosen next unique point, changes as step 2.251, if all unique points are all calculated and finished, enters above-mentioned steps 2.26.
6. a kind of image storehouse, reference mark matching process based on Image registration according to claim 2 is characterized in that, in the described step 2.29, the concrete steps of the first candidate matches peak point of multinomial model optimization unique point are as follows:
Step 2.291 is calculated the polynomial fitting coefficient, and the affined transformation model that uses in the calculating is as follows:
md x＝a 1+a 2×xd+a 3×yd，
md y＝b 1+b 2×xd+b 3×yd，
Wherein, md x, md yBe the parallax parameter of the first candidate matches peak point of unique point, xd=x-xc, yd=y-yc, x, y are characteristic point coordinates, xc, yc are the parallactic grid center point coordinate; With all R in the feature point set B＞1 unique point substitution affined transformation model, each feature point range one set of equations is calculated a by least square method 1, a 2, a 3, b 1, b 2, b 3Six affined transformation coefficients;
Step 2.292, optimize the first candidate matches peak point:
Steps A, calculated characteristics is put the deviate of the first candidate matches peak point, and concrete grammar is as follows:
Make updating mark f=0; Judge whether current unique point R in the feature point set B＞1, otherwise change step F, be then unique point substitution polynomial fitting to be calculated:
px＝a 1+a 2×xd+a 3×yd，
py＝b 1+b 2×xd+b 3×yd，
Wherein, px, py parallax value in polynomial fitting, calculating, xd=x-xc, yd=y-yc, x, y are characteristic point coordinates, xc, yc are the parallactic grid center point coordinate;
Calculation deviation value t again x, t y:
t x=| px-md x|, t y=| py-md y|; Wherein, md x, md yParallax parameter for the first candidate matches peak point of unique point;
Step B judges deviate: judge t x＞12 and t y＞12, be then to carry out step C, then change step F if not over to;
Step C, calculate the deviation of next candidate matches peak point: judging whether next candidate matches peak point of this unique point exists, is then to change step D; Otherwise commentaries on classics step e;
Step D upgrades the first candidate matches peak point; Concrete grammar is: make updating mark f=1; According to the parallax parameter of this candidate matches peak point, calculate new deviate t Xi, t Yi
Judge whether t Xi＜t xAnd t Yi＜t y, otherwise rotor step C is then to make t x=t Xi, t y=t Yi, as the first candidate matches peak point, its parallax parameter value is composed parallax parameter to unique point, and (dx dy), and makes the reliability value R of this unique point with this candidate matches peak point B=1, change step C;
Step e judges whether updating mark f=1, is then to change step F over to; Otherwise make this unique point reliability value R B=0, change step F;
Step F, judging characteristic point concentrates whether there is next unique point, and being then changes steps A, otherwise changes step 2.293;
Step 2.293 replacement parallactic grid grid value: each graticule mesh point value of parallactic grid (mx my) is:
Mx=a 1+ a 2* Sx+a 3* Sy, my=b 1+ b 2* Sx+b 3* Sy, wherein, Sx, Sy are the distance of this grid points to the graticule mesh center.
7. according to claim 2 or 3 described a kind of image storehouse, reference mark matching process, it is characterized in that in the described step 3, final matching results determines that method is as follows based on Image registration:
If the unique point of final entry in the step 2.5 is reliability value R B＞1, then calculate its image coordinate (x, y) to the Euclidean distance m of reference mark image center (x ', y '),
Figure FSA00000110834100081
Travel through all unique points, obtain minimum m value, promptly nearest apart from reference mark image center unique point is as the reference mark matching result.
8. an employing is based on the device of image storehouse, the reference mark matching process of Image registration, it is characterized in that, comprise geometric distortion cancellation module and the pyramid image matching module that links to each other with the geometric distortion cancellation module, the matching result acquisition module links to each other with described pyramid image matching module.
9. a kind of device that adopts based on image storehouse, the reference mark matching process of Image registration according to claim 8, it is characterized in that described geometric distortion cancellation module comprises parameter acquiring unit, regional acquiring unit subject to registration and the perspective transform parameter acquiring unit that links to each other successively.
10. a kind of device that adopts based on image storehouse, the reference mark matching process of Image registration according to claim 8, it is characterized in that described pyramid image matching module comprises matching parameter acquiring unit, first order matching unit, second level matching unit and the N level matching unit that links to each other successively.
CN201010161169.7A 2010-04-27 2010-04-27 Control point image database matching method based on image registration Expired - Fee Related CN101840517B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201010161169.7A CN101840517B (en) 2010-04-27 2010-04-27 Control point image database matching method based on image registration

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201010161169.7A CN101840517B (en) 2010-04-27 2010-04-27 Control point image database matching method based on image registration

Publications (2)

Publication Number Publication Date
CN101840517A true CN101840517A (en) 2010-09-22
CN101840517B CN101840517B (en) 2014-04-30

Family

ID=42743879

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201010161169.7A Expired - Fee Related CN101840517B (en) 2010-04-27 2010-04-27 Control point image database matching method based on image registration

Country Status (1)

Country Link
CN (1) CN101840517B (en)

Cited By (16)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103180874A (en) * 2010-10-27 2013-06-26 米其林企业总公司 Method for the pre-processing of a three-dimensional image of the surface of a tyre for use in the inspection of said surface
CN103542851A (en) * 2013-11-04 2014-01-29 东南大学 Underwater vehicle aided navigation positioning method based on underwater topography elevation database
CN103646389A (en) * 2013-03-26 2014-03-19 中国科学院电子学研究所 SAR slant range image match automatic extraction method based on geometric model
CN103761739A (en) * 2014-01-23 2014-04-30 武汉大学 Image registration method based on half energy optimization
CN103886338A (en) * 2014-02-13 2014-06-25 中国科学院光电研究院 Multispectral image matching method and system thereof
CN105160624A (en) * 2015-08-20 2015-12-16 中电科海洋信息技术研究院有限公司 Geographic information picture automatic registration method and apparatus
CN105323577A (en) * 2014-07-10 2016-02-10 三星电子株式会社 Multiple view image display apparatus and disparity estimation method thereof
CN106067024A (en) * 2015-04-01 2016-11-02 息科安宝 Feature point extraction device and method and the image matching system utilizing it
CN106157285A (en) * 2015-04-03 2016-11-23 株式会社理光 For selecting the method and system of the preferred value of the parameter group for disparity computation
CN106887016A (en) * 2017-03-09 2017-06-23 中国科学院遥感与数字地球研究所 A kind of automatic Relative matching method of the satellite sequence images of GF 4
CN107240130A (en) * 2017-06-06 2017-10-10 苍穹数码技术股份有限公司 Remote Sensing Image Matching method, apparatus and system
CN107784666A (en) * 2017-10-12 2018-03-09 武汉市工程科学技术研究院 The detection of terrain and its features three dimensional change and update method based on stereopsis
CN107945218A (en) * 2017-11-22 2018-04-20 中国资源卫星应用中心 The big distorted image matching process in edge
CN108763575A (en) * 2018-06-06 2018-11-06 湖南省第测绘院 Photo control point automatically selecting method based on photo control point database
CN112509042A (en) * 2020-11-27 2021-03-16 西安中科星图空间数据技术有限公司 Real-time positioning method and device based on-satellite control point library and storage medium
CN114567763A (en) * 2022-02-22 2022-05-31 湖南泊湾科技有限公司 Bilinear geometric correction method and system capable of adding points randomly

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20040037451A1 (en) * 2002-08-22 2004-02-26 Kim Tae Jung Image processing method for automatic image registration and correction
JP2004516533A (en) * 2000-09-22 2004-06-03 エイチアールエル ラボラトリーズ，エルエルシー Synthetic aperture radar and forward-looking infrared image superposition method
CN1987896A (en) * 2005-12-23 2007-06-27 中国科学院中国遥感卫星地面站 High resolution SAR image registration processing method and system
CN101598550A (en) * 2009-07-03 2009-12-09 武汉大学 Satellite stereoscopic image approximate nuclear line rapid rearrangement method based on projected footprint

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2004516533A (en) * 2000-09-22 2004-06-03 エイチアールエル ラボラトリーズ，エルエルシー Synthetic aperture radar and forward-looking infrared image superposition method
US20040037451A1 (en) * 2002-08-22 2004-02-26 Kim Tae Jung Image processing method for automatic image registration and correction
CN1987896A (en) * 2005-12-23 2007-06-27 中国科学院中国遥感卫星地面站 High resolution SAR image registration processing method and system
CN101598550A (en) * 2009-07-03 2009-12-09 武汉大学 Satellite stereoscopic image approximate nuclear line rapid rearrangement method based on projected footprint

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
《武汉大学学报信息科学版》 20031031 江万寿 等 航空影像特征匹配研究 第28卷, 第5期 2 *
《遥感应用》 20080229 徐硕 等 基于控制点影像库的控制点自动选取初探 , 2 *

Cited By (26)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103180874A (en) * 2010-10-27 2013-06-26 米其林企业总公司 Method for the pre-processing of a three-dimensional image of the surface of a tyre for use in the inspection of said surface
CN103646389A (en) * 2013-03-26 2014-03-19 中国科学院电子学研究所 SAR slant range image match automatic extraction method based on geometric model
CN103542851A (en) * 2013-11-04 2014-01-29 东南大学 Underwater vehicle aided navigation positioning method based on underwater topography elevation database
CN103542851B (en) * 2013-11-04 2016-03-23 东南大学 A kind of submarine navigation device assisting navigation localization method based on underwater topography elevation database
CN103761739B (en) * 2014-01-23 2016-09-28 武汉大学 A kind of Image registration method optimized based on half global energy
CN103761739A (en) * 2014-01-23 2014-04-30 武汉大学 Image registration method based on half energy optimization
CN103886338A (en) * 2014-02-13 2014-06-25 中国科学院光电研究院 Multispectral image matching method and system thereof
CN103886338B (en) * 2014-02-13 2018-03-27 中国科学院光电研究院 A kind of multispectral image matching process and system
CN105323577B (en) * 2014-07-10 2018-12-28 三星电子株式会社 Multi-view image shows equipment and its parallax estimation method
US10152803B2 (en) 2014-07-10 2018-12-11 Samsung Electronics Co., Ltd. Multiple view image display apparatus and disparity estimation method thereof
CN105323577A (en) * 2014-07-10 2016-02-10 三星电子株式会社 Multiple view image display apparatus and disparity estimation method thereof
CN106067024B (en) * 2015-04-01 2021-08-06 息科安宝 Feature point extraction device and method, and image matching system using the same
CN106067024A (en) * 2015-04-01 2016-11-02 息科安宝 Feature point extraction device and method and the image matching system utilizing it
CN106157285A (en) * 2015-04-03 2016-11-23 株式会社理光 For selecting the method and system of the preferred value of the parameter group for disparity computation
CN105160624A (en) * 2015-08-20 2015-12-16 中电科海洋信息技术研究院有限公司 Geographic information picture automatic registration method and apparatus
CN106887016B (en) * 2017-03-09 2020-07-03 中国科学院遥感与数字地球研究所 Automatic relative registration method for GF-4 satellite sequence images
CN106887016A (en) * 2017-03-09 2017-06-23 中国科学院遥感与数字地球研究所 A kind of automatic Relative matching method of the satellite sequence images of GF 4
CN107240130A (en) * 2017-06-06 2017-10-10 苍穹数码技术股份有限公司 Remote Sensing Image Matching method, apparatus and system
CN107784666A (en) * 2017-10-12 2018-03-09 武汉市工程科学技术研究院 The detection of terrain and its features three dimensional change and update method based on stereopsis
CN107784666B (en) * 2017-10-12 2021-01-08 武汉市工程科学技术研究院 Three-dimensional change detection and updating method for terrain and ground features based on three-dimensional images
CN107945218A (en) * 2017-11-22 2018-04-20 中国资源卫星应用中心 The big distorted image matching process in edge
CN107945218B (en) * 2017-11-22 2020-05-22 中国资源卫星应用中心 Edge large distortion image matching method
CN108763575A (en) * 2018-06-06 2018-11-06 湖南省第测绘院 Photo control point automatically selecting method based on photo control point database
CN108763575B (en) * 2018-06-06 2020-01-03 湖南省第一测绘院 Image control point automatic selection method based on image control point database
CN112509042A (en) * 2020-11-27 2021-03-16 西安中科星图空间数据技术有限公司 Real-time positioning method and device based on-satellite control point library and storage medium
CN114567763A (en) * 2022-02-22 2022-05-31 湖南泊湾科技有限公司 Bilinear geometric correction method and system capable of adding points randomly

Also Published As

Publication number Publication date
CN101840517B (en) 2014-04-30

Similar Documents

Publication Publication Date Title
CN101840517B (en) Control point image database matching method based on image registration
US10303966B2 (en) Method and system of image-based change detection
CN104574347B (en) Satellite in orbit image geometry positioning accuracy evaluation method based on multi- source Remote Sensing Data data
CN104091369B (en) Unmanned aerial vehicle remote-sensing image building three-dimensional damage detection method
CN102693542B (en) Image characteristic matching method
CN104637073B (en) It is a kind of based on the banding underground structure detection method for shining upon shadow compensation
CN103729853B (en) High score remote sensing image building under three-dimension GIS auxiliary damages detection method
CN103077527A (en) Robust multi-source satellite remote sensing image registration method
CN104156957B (en) Stable and high-efficiency high-resolution stereo matching method
CN104299228B (en) A kind of remote sensing image dense Stereo Matching method based on Accurate Points position prediction model
CN102609918A (en) Image characteristic registration based geometrical fine correction method for aviation multispectral remote sensing image
CN102654902A (en) Contour vector feature-based embedded real-time image matching method
CN105387846B (en) The ortho-rectification method and system of satellite image
CN104599281B (en) A kind of based on the conforming panorama sketch in horizontal linear orientation and remote sensing figure method for registering
CN104050666A (en) Brain MR image registration method based on division
CN103235810B (en) Remote sensing image reference mark data intelligence search method
CN106023230A (en) Dense matching method suitable for deformed images
CN102081733A (en) Multi-modal information combined pose-varied three-dimensional human face five-sense organ marking point positioning method
CN102903109A (en) Integrated partition registering method of optical image and synthetic aperture radar (SAR) image
CN101957916A (en) Method for extracting affine invariant feature of image by using M-band wavelet
Galeon Estimation of population in informal settlement communities using high resolution satellite image
CN107784666A (en) The detection of terrain and its features three dimensional change and update method based on stereopsis
EP4078534A1 (en) Maps comprising covariances in multi-resolution voxels
CN104463895B (en) A kind of earth's surface monitoring image processing method based on SAR
Noaje et al. Environmental changes analysis in Bucharest City using Corona, SPOT HRV and IKONOS images

Legal Events

Date Code Title Description
C06 Publication
PB01 Publication
C10 Entry into substantive examination
SE01 Entry into force of request for substantive examination
C14 Grant of patent or utility model
GR01 Patent grant
CF01 Termination of patent right due to non-payment of annual fee

Granted publication date: 20140430

Termination date: 20150427

EXPY Termination of patent right or utility model